Genes Contributing to Domestication of Rice Seed Traits and Its Global Expansion

Asian rice (Oryza sativa) and African rice (Oryza glaberrima) are separately domesticated from their wild ancestors Oryza rufipogon and Oryza barthii, which are very sensitive to daylength. In the process of domestication, some traits that are favorable for the natural survival of wild rice such as seed dormancy and shattering have become favorable ones for human consumption due to the loss-of-function mutations in the genes that are underlying these traits. As a consequence, many genes that are related to these kinds of traits have been fixed with favorable alleles in modern cultivars by artificial selection. After domestication, Oryza sativa cultivars gradually spread to temperate and cool regions from the tropics and subtropics due to the loss of their photoperiod sensitivity. In this paper, we review the characteristics of domestication-related seed traits and heading dates in rice, including the key genes controlling these traits, the differences in allelic diversity between wild rice and cultivars, the geographic distribution of alleles, and the regulatory pathways of these traits. A comprehensive comparison shows that these genes contributed to rice domestication and its global expansion. In addition, these traits have also experienced parallel evolution by artificial selection on the homologues of key genes in other cereals.


Introduction
A suite of common traits is selected during the domestication of crops, which are collectively known as a "domestication syndrome". These traits make domesticated species different from their wild ancestors. In cereals, these traits include morphological traits such as larger grains, a loss of seed dispersal, increased apical dominance, more determinate growth, and physiological traits including seed dormancy, seed shattering, bitter substances in edible structures, photoperiod sensitivity, and synchronized flowering [1]. The seed-related traits of dormancy, seed shattering, and grain size determine which kinds of grain humans can choose for domestication, and the photoperiod-sensitive heading date limits the optimal growing region of rice. Rice is a model crop plant due to its small genome size, high-quality genome sequence, and efficient transformation [2]. In the last two decades, many genes that are related to domestication and global expansion have been isolated in rice. In this review, we summarize the genes controlling seed dormancy, seed shattering, grain size, and flowering time, which contribute to rice domestication and its global expansion. In addition, the homologues in other crops are also discussed to verify parallel evolution during domestication.

Seed Shattering in Rice
Wild rice seeds fall off after maturity to ensure their natural propagation in the natural environment. However, seed shattering causes yield loss for domesticated crop plants during harvest. The nonshattering trait is likely to undergo strong selection early in domestication and further facilitate the fixation of other domestication characters. Hence, the loss of seed shattering is considered to be direct ecological evidence for wild rice domestication.

Seed Shattering Genes Identified from Natural Variation
In rice, seed shattering is regulated by the formation of an abscission zone (AZ), which is composed of several layers of small and dense cytoplasmic cells in the joint between the lemma and pedicel [3]. Several shattering quantitative trait loci (QTLs) have been identified including SH4, encoding an Myb3 transcription factor, and qSH1, encoding a BELL1-like homeodomain protein (Table 1) [4][5][6]. The mutated alleles qsh1 and sh4 cause seed non-shattering owing to the absence of abscission layer formation ( Table 1). The SHAT1 gene, which encodes an APETALA2 transcription factor, is identified from induced mutant through a 60Co γ-ray radiation and is required for seed shattering through specifying AZ development in rice. The expression of SHAT1 in the abscission layer is positively regulated by SH4, whose mutation results in incomplete development and the partial functioning of the AZ. qSH1 functions downstream of SHAT1 and SH4, and is involved in a positive feedback loop with SHAT1 and SH4 through the maintenance of SHAT1 and SH4 expression in the AZ, thereby promoting AZ differentiation [7]. Os03g0407400 G protein γ subunit C/A substitution in exon [11,12] qLGY3/OsLG3b LOC_Os03g11614 Os03g0215400 MADS transcription factor Six SNPs in exon [13,14] GW5/qSW5 LOC_Os05g09520 Os05g0187500 Plasma membrane 1212-bp or 950-bp deletion in the promoter [15][16][17][18] Bp: Base pair, SNP: Single nucleotide polymorphism; G/T: G substitutes for T; C/A: C substitutes for A; Myb-like: Myeloblastosis like; bHLH: Basic helix-loop-helix; MADS: MCM1, AGAMOUS, DEFICIENS, SRF.

Mutated Alleles Led to Seed Non-Shattering Domestication
Sh4 was detected using an F2 population derived from an indica-type cultivar and the wild rice progenitor Oryza nivara (an annual form of Oryza rufipogon). The nucleotide substitution guanine/ thymine (G/T) in the first exon of SH4 results in a mutated allele carried by all of the cultivars compared with the alleles in wild rice. The SH4 phylogeny, together with neutrality tests and coalescent simulations, suggested that sh4 had a single origin and was fixed to have a mutated allele by artificial selection during domestication [4,5,19], indicating its vital role in rice domestication. However, a near isogenic line carrying sh4 alleles showed strong seed-shattering behavior in the wild rice genetic background and weedy rice, indicating that other genes redundantly regulate abscission layer formation [20][21][22][23]. The interaction between the sh4 and qSH3 loci inhibits the formation of the abscission layer in rice [20][21][22]24,25]. The effect of qSH3 on seed shattering was weaker than that of qSH1 or sh4 when evaluated in the genetic background of cultivated rice [24,25]. qSH1 was detected using a mapping population derived from the japonica cultivar Nipponbare and indica cultivar Kasalath. The single nucleotide polymorphism (SNP) (G/T) in the 5' regulatory region of qSH1 was conserved in japonica and highly associated with shattering in the temperate japonica subspecies, and has not been introgressed into indica subspecies [6,19].

The Parallel Evolution of the Non-Shattering Trait in Cereal Crops
Some seed shattering genes are identified from Asian rice including SH4 and qSH1, and some other genes are identified in African rice, including GL4 and ObSH3 [4][5][6]26,27]. GL4, encoding an Myb3 transcription factor, was identified in Oryza barthii, which is the ancestor of African cultivars (Oryza. glaberrima). A cytosine/thymine (C/T) substitution in the GL4 gene resulted in a premature stop codon and led to small seeds and the loss of seed shattering [26]. ObSH3, encoding a YABBY transcription factor, was identified from an F 2 population derived from a cross between the African wild rice accession W1411 and the non-shattering African cultivar IRGC104165. Most accessions with both mutations gl4 and obsh3 occur in arid regions north of 11 • N in African [26,27]. Their orthologues in either African rice or Asian rice have the conserved functions and have been subjected to selection. For example, OgSH1 and OgSH4, the orthologs of the Asian rice-shattering genes OsSh1 and SH4, also play important roles in controlling seed shattering in African rice. Ossh4 and Ogsh4 were selected in parallel during the domestication of African and Asian rice, respectively [4,5,27,28]. Therefore, artificial selection on common shattering genes resulted in non-shattering in Asian and African cultivars. Meanwhile, seed non-shattering due to the loss of the abscission layer also exists in other crops. Sh1, a homologue of OsSH1 in sorghum, also regulates seed shattering [29]; ZmSh1-1, ZmSh1-5.1, and ZmSh1-5.2, the homologues of ObSH3 in maize, have been verified to control seed shattering by QTL mapping with a large mapping population [27,29]. The Q gene, the homologue of SHAT1 in wheat, encodes a member of the APETALA2 family of transcription factors and confers the free-threshing character [30,31]. The same regulatory point mutation in rice Sh1 was also identified in an Sh1 orthologue gene in Brassica, and is responsible for the seed dispersal structures produced by natural selection [32]. Taken together, these findings indicate that parallel selection on non-shattering exists during sorghum, rice, and maize domestications.

Seed Dormancy in Rice
Seed dormancy is a special period in the plant life cycle when seed germination is unable to proceed under a combination of environmental factors that are normally suitable for the germination of non-dormant seeds. Dormancy is a mechanism in wild species to prevent germination during unsuitable external conditions. Excessive seed dormancy is not a desired trait in crops. Some degree of dormancy is desired to prevent pre-harvest sprouting. Hence, domesticated crops have undergone selection against dormancy. Abscisic acid (ABA) and gibberellin (GA) are two major regulators of seed dormancy and germination. Abscisic acid positively regulates dormancy induction and maintenance, while GA promotes seed germination [33,34].

Seed Dormancy Genes Identified from Natural Variation
Many QTLs affecting seed dormancy or germination-related traits have been mapped in rice [35][36][37][38][39][40][41], but only a few QTLs have been cloned in rice (Table 1), such as Sdr4, encoding a novel protein; qSD7-1/Rc, encoding a bHLH transcription factor; and qSD1-2, encoding OsGA20ox2 [8][9][10]42]. Sdr4 acts as a seed dormancy-specific regulator that is under the control of OsVP1, which is a global positive regulator of seed maturation through its effect on ABA signaling and the regulation of the expression of OsDOG1L-1, which is a positive regulator of dormancy in rice [8]. qSD7-1/Rc controls dormancy and pigment traits by regulating ABA and the flavonoid biosynthetic pathways, respectively [9]. However, qSD1-2 regulates seed dormancy by controlling the seed GA level [10]. A recent study demonstrated that the antagonistic relationship between the ABA and GA metabolic pathways regulates the switch of cereal seeds between dormancy and germination [43].

Mutated Alleles Led to Seed Dormancy Domestication
The 18-bp (base pair) direct repeat sequence variation resulting from a double-strand cleavage and repair event in Sdr4 substantially contributes to the differences in seed dormancy between japonica (Nipponbare) and indica (Kasalath) cultivars. A sequence analysis of 59 cultivars and 46 accessions of O. rufipogon revealed that the Sdr4-n sequence that causes reduced dormancy was not found in any wild rice accessions, and that the Sdr4-n in indica cultivars is introgressed from japonica rice. Sdr4-n appears to have been produced through at least two mutation events from the closest O. rufipogon allele among the examined accessions. Sdr4-k and Sdr4-k in indica were inherited from these subgroups of the wild ancestor [8]. qSD7-1/qPC7 controls the seed dormancy/pericarp color in weedy red rice. The dormancy-enhancing alleles qSD7-1/qPC7 was differentiated into two groups that are generally associated with the tropical and temperate ecotypes of weedy rice. qSD7-1/qPC7 may contribute the most to weed adaptation [9]. A loss-of-function mutation in qSD1-2 enhances seed dormancy and results in semi-dwarfism, which has been used to develop high-yield, semi-dwarf varieties worldwide. The sd1 mutant originally occurred in an O. rufipogon population and in weedy rice. The allelic distribution of qSD1-2/OsGA20ox2 was found to be associated with the subspeciation of indica and japonica rice [44]. However, there is no evidence that the primitive indica-specific and japonica-specific alleles are functionally differentiated at the presumably domestication-related locus of qSD1-2/OsGA20ox2 [35].

The Parallel Evolution of Seed Dormancy in Cereals
Recently, many QTLs affecting seed dormancy or germination-related traits have been identified in plant species such as barley [45][46][47] and wheat [40,[48][49][50]. However, most seed dormancy QTLs have not been finely mapped. Currently, it is difficult to infer the identities of these QTLs. Based on synteny analysis, HvGA20ox has been suggested as the candidate of a seed dormancy QTL in barley [47]. The orthologues of OsSdr4 in wheat, namely TaSdr, in which a single-nucleotide mutation causes a distinct phenotype in seed germination, have been demonstrated to be key regulators of pre-harvest sprouting in wheat [51]. Seed dormancy genes are probably under parallel selection in cereals, which needs to be further demonstrated by testing the function of more homologues of seed dormancy genes in other cereal crops in the future.

Grain Size in Rice
Plant seeds are major sources of human nutrition and are the major means of crop propagation. The grain size of wild rice is significantly smaller than that of cultivated rice [1]. Early in domestication, farmers preferred to select larger seeds to increase grain yield and obtain more food during domestication [52]. Therefore, seed size has always been subjected to selection.

Mutated Alleles Contributing to Domestication
GS3 contains four putative domains, and the organ size regulation (OSR) domain is necessary and sufficient for its function as a negative regulator [11,12]. GS3 has undergone positive selection in cultivars [75,76]. A cytosine/adenine (C/A) mutation in the second exon is associated with large grain size and classifies global rice collections into long-grain and short-grain groups. C/A variation exists in the wild rice gene bank [75,77,78], and the C/A variation in wild rice may have resulted from the gene flow from cultivated rice. However, this mutation does not play a similar role in wild rice [77]. The short-grain alleles of GS3 have multiple independent origins, because farmers and early breeders imposed artificial selection, favoring short seeds [79]. The long-grain allele gs3 likely originated from a japonica-like ancestor, and was subsequently introduced into indica by introgression [77]. However, a haplotype network and phylogenetic analyses showed that the japonica and indica haplotypes evolved independently [80].
qLGY3/OsLG3b encodes the MADS-domain transcription factor OsMADS1, and it regulates grain size by interacting with the Gγ subunits GS3 and DEP1. Six SNPs in the OsLG3b region led to alternative splicing, which was associated with grain length, resulting in increases in both the grain quality and yield potential of rice [13,14]. Haplotype analysis revealed that the long-grain allele of OsLG3b might have arisen after the domestication of tropical japonica, and then spread to the subspecies indica or temperate japonica by natural crossing and artificial selection, which is similar to the events related to the GS3 gene that lead to the improvement of tropical japonica [14].
GW5/qSW5 encodes a plasma membrane-associated protein with IQ calmodulin-binding motifs, and is a novel positive regulator of brassinosteroid (BR) signaling that controls grain width and weight through the proteasomal degradation pathway [15][16][17][18]. A 1212-bp deletion (DEL2) in japonica varieties and a 950-bp deletion (DEL1) in indica varieties in the promoter region of qSW5 are confirmed to be the causal mutations that led to increases in both grain width and grain weight, and have strong correlations with grain width in rice. Both the DEL1 and DEL2 deletions likely originated in different wild rice accessions during rice domestication, and were enriched by artificial selection, as well as the propagation of cultivation and natural crosses, and eventually became widely utilized by rice breeders [15][16][17][18]. A nucleotide diversity analysis showed that qSW5 has high nucleic acid polymorphism in both cultivated and wild rice populations, and it has been subjected to positive selection for genetic improvement [81].

The Parallel Evolution of Grain Size in Cereals
The GS3-homologous gene ZmGS3 in maize contains the same conserved domain as GS3. Correlation analysis shows that the SNP located in the fifth exon is significantly correlated with the kennel length of maize [82]. The TaGW2-6A of a GW2 homologue in wheat and the ZmGW2-CHR4 and ZmGW2-CHR5 of GW2 homologues in maize are highly associated with grain width and grain weight in a large germplasm collection [83,84]. The orthologue of rice GS5 in wheat, TaGS5-3A-T, is significantly associated with larger grain size and a higher thousand-kernel weight [85]. In contrast to the evolutionary model of GS5 in rice, TaGS5-3A-T is positively selected in Chinese wheat breeding, and undergoes selection during each polyploidization event [85,86]. Among the 15 seed size genes previously identified to be under selection in rice or maize, 12 orthologues in sorghum have been under selection during domestication [87]. The major grain size genes in rice and their homologues in other cereal crops have large effects on grain weight; therefore, they have been easily selected based on trait performance. Therefore, grain size has undergone parallel evolution in cereals.

Flowering and Adaptation in Rice Expansion
Rice flowering time is determined by internal genetic factors and external environmental factors such as daylength, temperature, drought, nutrients, and biotic stresses. The molecular mechanism of rice flowering has been well characterized and has been recently reviewed [88]. Many genes are included in the review, but here we mainly focused on the genes that have been isolated from natural variations and are related to rice adaptation. O. rufipogon is mainly in a limited tropical region nearby and has been domesticated to the cultivar O. sativa in the Yangtze River valley region in China [78,89]. After a long process of domestication and improvement, rice has been successfully grown worldwide as a global crop, indicating that the rich variation in heading date genes has allowed cultivated rice to adapt to different environments, especially the changing daylength at different latitudes [90]. Rice is a short-day plant. Rice flowering is repressed when the daylength is longer than 13.5 h [91]. In tropical regions, the daylength is less than 13.5 h, and the daily temperature is high, which ensures that rice can grow all year. However, in other regions, such as temperate regions, rice only grows in the short and warm summer, when the daylength is long. Therefore, during rice expansion, cultivars have gradually adapted to the long daylength at high latitudes. Here, we will review the genes that are responsible for rice adaptation to various daylengths, trace the nucleotide changes, and try to reveal how cultivars expanded into diverse regions.

Heading Date Genes Identified from Natural Variation
Cultivars have abundant variation in flowering time [92], indicating that many heading date genes are responsible for rice flowering. Dozens of genes have been mapped for heading date in rice [93]. Eighteen QTLs (Hd1-Hd18) were detected with the different populations that have been derived from crosses between Nipponbare and Kasalath and between Koshihikari and Hayamasari, and most of the QTLs have been cloned [94][95][96][97]. Hd1, the first cloned flowering gene from natural variation in rice, is the orthologue of CONSTANT, which encodes a transcription factor with a zinc-finger domain and a CONSTANS, CO-like, and TOC1 (CCT) domain. This gene is a bifunctional regulator that promotes flowering in short days and delays flowering in long days [98]. Hd2, which is also named Ghd7.1/OsPRR37/DTH7, encodes a transcription factor with a pseudoreceiver domain and a CCT domain. This gene delays flowering under long days [99][100][101]. qHd3 includes two heading date genes, namely Hd3a and Hd3b (Hd17/OsELF3/EF7) [102][103][104]. Hd3a encodes one of the mobile flowering signal florigens in rice, and RFT1 encodes another florigen. Both genes belong to the phosphatidylethanolamine-binding protein (PEBP) gene family [105][106][107][108]. Hd3b, which is also named OsELF3, Hd17, and EF7, is the homologue of ELF3, which is a component of the Arabidopsis circadian clock [104]. Hd4/Ghd7, encoding a transcription factor with a CCT domain, delays flowering on long days [109]. Hd5/Ghd8/DTH8/LHD1 is a member of the heme-associated proteins 3 (HAP3) family [110][111][112]. Hd6 and Hd16/EL1/EF7 encode the subunits of casein kinase II and casein kinase I separately [113][114][115]. Hd9/DTH3/OsMADS50, the homologue of SOC1 in rice, is a member of the MADS-box gene family [116]. Hd18 encodes an amine oxidase domain-containing protein and is the homologue of Arabidopsis flowering locus D [97]. With other bi-parentally derived populations, additional heading date QTLs have been identified and cloned such as Ehd1, DTH2, and Ehd4 [117][118][119]. Ehd1 encodes a B-type response regulator and promotes flowering independently of Hd1. DTH2 also encodes a transcription factor with a zinc-finger domain and CCT domain, and delays flowering on long days. Ehd4 encodes a CCCH-type zinc finger transcription factor and promotes flowering. Interestingly, when QTL analyses were performed using F2 populations from crosses between Koshihikari and 12 cultivars originating from various regions in Asia, most of the major QTLs were identified, including Hd1, Hd6, Hd16, Hd17, Ghd7, Ghd7.1, Ghd8, Hd3a, and RFT1 [120][121][122]. In addition, many of the minor QTLs were identified with advanced populations [122]. Genome-wide association analysis studies on the basis of diverse germplasm collections identified some of the flowering genes that were detected in bi-parental mapping populations [123][124][125]. These results strongly suggest that these genes largely determine the variations in the heading date of rice.

Regulatory Networks of Flowering
Combined with the latest research findings, a draft regulatory network of how these QTLs/genes coordinate to determine rice flowering time is suggested (Figure 1). The key regulators of flowering are the florigens Hd3a and RFT1, which are expressed in leaves, and then move to the shoot to accelerate flowering [105][106][107][108]. Hd1 only triggers the expression of florigen genes directly in leaves, but it can switch to repress the expression of florigen genes when Ghd7 and Ghd8 are present in long days [126][127][128][129][130]. Ehd1 also triggers the expression of florigen genes independently of Hd1; moreover, it is regulated by Ghd7, Ghd8, Ghd7.1, Ehd4, DTH2, DTH3, and other flowering genes [101,109,111,112,116,118]. In addition, other regulators that are dependent on these major genes have been identified as contributing to flowering. Hd17 is a component of the circadian clock that delays flowering, depending on Hd1 [104]. Hd6 and Hd16 encode kinase and directly phosphorylate GHD7 and GHD7.1, and interact genetically with Ghd7, Ghd7.1, Ghd8, and Hd1 to further delay flowering [113,114,131]. Even more extreme, there are two combinations that lead to non-flowering in natural long days. One is the combination of functional Hd1, Ghd7, and Ghd8 in Zhenshan97, while another includes non-functional ehd1 and rft1 in both the Nona Bokra background and a recombinant inbred line derived from Guangluai 4 and Taichung 65 [126,132,133]. Taken together, these studies show that cultivars with different combinations of known flowering genes can exhibit extensive variation in the heading date, from earlier flowering to extremely later flowering or even non-flowering on long days. On short days, only a few genes such as Hd1, Ehd4, DTH3, and Hd18 still promote flowering as they do on long days, while most others have weak effects or even no effect such as Ghd7, Ghd7.1, Ghd8, Hd17, and DTH2 [100,104,109,111,[116][117][118][127][128][129].  [101,109,111,112,116,118]. In addition, other regulators that are dependent on these major genes have been identified as contributing to flowering. Hd17 is a component of the circadian clock that delays flowering, depending on Hd1 [104]. Hd6 and Hd16 encode kinase and directly phosphorylate GHD7 and GHD7.1, and interact genetically with Ghd7, Ghd7.1, Ghd8, and Hd1 to further delay flowering [113,114,131]. Even more extreme, there are two combinations that lead to non-flowering in natural long days. One is the combination of functional Hd1, Ghd7, and Ghd8 in Zhenshan97, while another includes non-functional ehd1 and rft1 in both the Nona Bokra background and a recombinant inbred line derived from Guangluai 4 and Taichung 65 [126,132,133]. Taken together, these studies show that cultivars with different combinations of known flowering genes can exhibit extensive variation in the heading date, from earlier flowering to extremely later flowering or even non-flowering on long days. On short days, only a few genes such as Hd1, Ehd4, DTH3, and Hd18 still promote flowering as they do on long days, while most others have weak effects or even no effect such as Ghd7, Ghd7.1, Ghd8, Hd17, and DTH2 [100,104,109,111,[116][117][118][127][128][129].

Diverse Alleles of Flowering Genes in Wild Rice and Cultivars
Oryza rufipogon has limited genetic and nucleotide diversity for a partial outcrossing species [89,134]. Here, we summarize the functional diversities of known flowering genes in wild rice and cultivars ( Table 2). Some key flowering genes have already diversified into functional (strong) and non-functional alleles in wild rice such as Hd1, Hd6, Hd16, and RFT1, and some genes such as Ehd4, Hd17, and Hd3a have generated weak alleles. However, no non-functional alleles have been identified for the strong photoperiod sensitivity genes Ghd7, Ghd7.1, Ghd8, DTH2, and DTH3 in wild rice. These results indicate that the functional mutated alleles of some flowering genes exist in wild rice, but only in rare wild rice accessions. In cultivars, all of the flowering genes have several kinds of mutant alleles with a weak effect or no effect, in addition to the pre-existing alleles in wild rice. The newly generated alleles are enriched in high-latitude regions, because they are artificially selected to be grown under

Diverse Alleles of Flowering Genes in Wild Rice and Cultivars
Oryza rufipogon has limited genetic and nucleotide diversity for a partial outcrossing species [89,134]. Here, we summarize the functional diversities of known flowering genes in wild rice and cultivars ( Table 2). Some key flowering genes have already diversified into functional (strong) and non-functional alleles in wild rice such as Hd1, Hd6, Hd16, and RFT1, and some genes such as Ehd4, Hd17, and Hd3a have generated weak alleles. However, no non-functional alleles have been identified for the strong photoperiod sensitivity genes Ghd7, Ghd7.1, Ghd8, DTH2, and DTH3 in wild rice. These results indicate that the functional mutated alleles of some flowering genes exist in wild rice, but only in rare wild rice accessions. In cultivars, all of the flowering genes have several kinds of mutant alleles with a weak effect or no effect, in addition to the pre-existing alleles in wild rice. The newly generated alleles are enriched in high-latitude regions, because they are artificially selected to be grown under long-day conditions due to their weak or lack of photoperiod sensitivity [104,117]. Both Ghd7 and Ghd7.1 are sensitive to photoperiod. The alleles with strong effects are preserved from wild rice, but the indica and japonica alleles independently originated from different wild rice accessions [100,135], which indicates that indica-japonica differentiation has already occurred in wild rice. This notion is also supported by an evolutionary analysis of a major reproductive barrier regulator, S5 [136]. Weak alleles or non-functional alleles of Ghd7 and Ghd7.1 were then generated in japonica and indica in parallel. Thus, the retention of their pre-existing genetic variants in ancestral species and the acquisition of mutations after domestication have increased the natural variation in the heading date.
An analysis of the eco-geographical distribution patterns of flowering genes showed that flowering repressors are present in low-latitude regions at high frequencies, while the activators (except Hd1) are present in high-latitude regions at high frequencies. As an important flowering signal integrator, the exception-Hd1-has an eco-geographical distribution pattern that is well explained by its bifunctionality. When Hd1 is combined with the non-functional alleles ghd7 and ghd8, Hd1 promotes flowering under either long-day or short-day conditions [128]. When Hd1 is combined with Ghd7 and Ghd8, Hd1 interacts with the repressors of Ghd7 and Ghd8 under long-day conditions, and in turn greatly delays the flowering of this genotype, which is mainly grown in tropical regions. In addition, the interactions between Hd6, Hd16, and Ghd7, Ghd8, Ghd7.1, Hd1 significantly increase the effects of these genes on the heading date [114,137,138]. These genes likely work together and may be the components of a large complex. Therefore, these interactions should be taken into consideration when developing a cultivar for a local region. With the help of humans/breeders, cultivars with different gene combinations are grown in optimized ecological regions, including high-latitude and low-latitude regions, to generate the maximum rice product [126]. However, for the flowering activators such as RFT1, the defective allele would be limited to the tropical region, and functional alleles would move to northern regions or the regions with double-cropping seasons such as those where early rice is grown, where the daylength is long throughout the cropping season [132,133]. Similarly, the functional Hd18 with a minor effect is present in Northeastern China at a higher frequency [139].

The Parallel Evolution of Key Flowering Genes in Cereals
Plants are classified into three types according to their responses to photoperiods: short-day plants such as rice, maize, and sorghum; long-day plants such as wheat and barley; and day-neutral plants such as tomato [141]. When grown under the same daylength conditions, short-day and long-day plants show opposite responses. However, accessions with different photoperiod response-related genes exhibit various photoperiod responses. We have described how Asian rice expands into different latitudes from its limited region of origin. The other crops such as wheat, maize, and sorghum also adapt to local photoperiod conditions and consequently expand to different latitudes.
The major determinant of long-day response in barley has been cloned as photoperiod-H1 (Ppd-H1) in a colinear region of Ghd7.1. In spring barley, an amino acid change in the CCT domain caused a reduced photoperiod response, which extended the growth period of barley, allowing it to adapt to the long growing seasons and produce higher yields in Western Europe and North America [142]. Different mutations in SbPRR37, the Ghd7.1 homologue in sorghum, also reduce the photoperiod sensitivity and cause sorghum to flower earlier, which is critical for the cultivation of this tropical crop in temperate regions worldwide [143]. BvBTC1, the homologue of Ghd7.1 in sugar beet (Beta vulgaris), is a master switch distinguishing annual from biennial. The loss-of-function BvBTC1 allele confers a reduced photoperiod sensitivity, later flowering, and bienniality [144]. Taken together, these findings indicate that Ghd7.1 and its homologues are key factors in expanding the cultivation of cereals. However, no identical mutant alleles were found in different crops, indicating that mutations occurred after species differentiation, but not from the common ancestor. Ghd7 is another important major determinant of photoperiod response and adaptation in rice. ZmCCT9 and ZmCCT10 are homologues of Ghd7 in maize. Insertions of transposable elements in the promoters of both genes caused a change in mRNA expression, and thus contributed to maize adaptation to higher latitudes after maize was domesticated from southern Mexico [145]. Mutations of EAM8, the homologue of Hd17, facilitate adaptation to a short growing season in barley and expand the geographic range of this species [146]. In sorghum, besides Ma1/SbPRR37 and EAM8, Ma6, which is the homologue of Ghd7, and FT, which is the homologue of Hd3a and RFT1 in sorghum, also regulates photoperiod flowering, and collectively with other genes contribute to its adaptation to diverse environments [147]. In addition, the parallel domestication of Hd1 was elucidated in sorghum, foxtail millet, and rice [148]. Thus, the photoperiod response genes Ghd7.1, Ghd7, Hd1, ELF3, Hd3a, and their homologues underwent parallel selection, and the loss of function of these genes helped crops extend their ranges into higher latitudes from the original range in the tropics or subtropics.

Conclusions
As described above, many important genes contribute to domestication in rice by loss of function. Moreover, many homologues in other cereals have conserved functions, which indicates that parallel evolution plays an important role in the "domestication syndrome". Parallel evolution provides us with the chance to examine the function of homologues in other cereals once a major gene for such traits has been identified. In general, domestication syndrome genes are divided into two types according to their distribution in cultivars and wild rice. The genes of type one control prostrate growth (not included in this review), seed dormancy, and seed shattering. Loss-of-function mutations in these genes are required for rice domestication, because these mutations allow cultivars to grow more easily and produce more grains initially. In contrast, the genes of type two have diverse functional alleles in cultivars, but some strong functional alleles in cultivars come directly from wild rice, and weak or non-functional alleles are generated in cultivars such as genes for grain size and flowering time. These genes contribute to rice domestication but play more important roles in rice global expansion and genetic improvement to develop modern rice. In brief, the domestication of wild traits such as prostrate growth, seed dormancy, and seed shattering is the sudden result of loss-of-function mutations, and adaptive traits determining rice expansion such as photoperiod sensitivity are the continuing results of the mutations from strong alleles to weak alleles, and then to non-functional alleles (Figure 2).

Conflicts of Interest:
The authors declare no conflict of interest.

Conflicts of Interest:
The authors declare no conflict of interest.