Introduction

China, with abundant rice genetic resources, is one of the centers of origin of Oryza sativa L. (Oka 1988), and one of centers of genetic diversity. China is exclusive in having developed large and nearly equal numbers of indica and japonica varieties. There were 56,020 accessions of cultivated rice in the ex situ germplasm collections of China until 1993 (ICGR CAAS 1996), 90.2% of which (50,526 accessions) were landraces, cultivated varieties (cultivars) grown and preserved by traditional farmers (Zeven 1998). Landrace varieties contain greater genetic diversity than elite cultivars (or commercial cultivars) and represent an intermediate stage in domestication between wild and elite cultivars (Londo et al. 2006). The abundant variation provides an important reservoir of genetic diversity and potential sources of beneficial alleles for rice breeding and improvement.

It is generally believed that the two primary subspecies of rice, indica and japonica, are the products of separate domestications from the ancestral species, O. rufipogon Griff. The evidence came from studies of biochemical traits (Second 1982) and hybrid sterility (Kato et al. 1928), subsequently supported by molecular analyses (Cheng et al. 2003), especially some new results on the evolutionary analysis of some domestication-related genes (Sweeney et al. 2007; Li et al. 2006; Konishi et al. 2006; see also review article by Kovach et al. 2007). However, studies using large numbers of cultivars indicated that some varieties do not belong to either of the two types, but they were generally related to differentiation of seasonal (Zhang et al. 2007a, b), soil-watery (Zhang et al. 2007a, b), or geographic ecotypes (Glaszmann 1987; Sano and Morishima 1992). Cai and Morishima (2000) even suggested that geographical differentiation might precede seasonal ecotype differentiation to produce groups like aman, aus, and boro. Using model-based structure analysis, Garris et al. (2005) divided O. sativa into five distinct groups, viz. indica, aus, aromatic, temperate japonica, and tropical japonica. This indicated that, in addition to the apparent differentiation between indica and japonica, other genetic structures or taxa, have evolved, but still have not been clearly identified and recognized. Therefore, it was of interest to (1) investigate the genetic structure of Chinese cultivated rice in relation to the process of intra-specific differentiation, and further, to understand the dynamics of intra-specific differentiation of cultivated rice, and (2) to gain information able to guide plant breeders in choosing parents for crossing between subspecies or ecotypes, and thus providing a more rational basis for expanding the gene pool and for identifying materials that harbor alleles of value for rice improvement.

The purpose of this study was (1) to investigate the population structure of 3,024 rice landrace accessions using SSRs, and (2) to examine the evolutionary relationships among intra-specific populations with emphasis on the intra- and inter-subspecific differentiation.

Materials and methods

Plant materials and predefined populations

The primary core collection of Chinese rice landraces, comprising 3,024 accessions, was used in this study. This collection was sampled from the basic collection of 50,526 accessions of rice landraces collected from all 32 provinces of China and represented 95% of the diversity in the basic collection at the morphological level (Li et al. 2003). In the database of Chinese Crop Germplasm Information System (http://icgr.caas.net.cn/cgris_english.html), all the accessions of rice landraces were documented by Ting’s five-level hierarchical taxonomy, i.e., two subspecies (indica or hsien versus japonica or keng), three photothermic ecotypes or seasonal ecotypes (early sown, medium sown, or late sown), two soil water regime ecotypes (hereafter called soil-watery ecotypes for simplicity) (lowland or upland), two endosperm types (waxy or non-waxy), and cultivars (Ting 1957). Studies of rice landraces from Yunnan and Guizhou have shown that the population structure of cultivated rice does not relate to endosperm types. According to the documented identities based on the first three levels of Ting’s taxonomy, 3,024 accessions were grouped into 14 populations, which we called predefined populations. Table 1 shows the distribution of cultivars in predefined populations.

Table 1 Distribution of predefined taxonomic identities for 3,024 accessions according to the first three levels of Ting’s taxonomic system
Full size table

DNA extraction and SSR genotyping

Thirty-day seedlings of 10–20 individuals for each accession were collected and ground in liquid nitrogen. DNA was extracted from the ground tissues using the CTAB method. Thirty-six SSRs (three on each of the 12 chromosomes, Table S1) were randomly selected to analyze population structure and genetic diversity. SSR loci are particularly useful for the study of population structures and demographic histories of domesticated species in natural populations (Goldstein and Schlotterer 1999) because of their genomic abundance, conservation of distinctive flanking sequence across closely related species, apparent selective neutrality, and high variability, that facilitate the detection of greater diversity than by using equal numbers of RFLP, AFLP, or SNP loci (Akagi et al. 1997).

The volume of the PCR reaction system was 15 μl, containing 20 ng DNA template. The PCR procedure followed Panaud et al. (1996) with slight modifications: (1) initial denaturation at 94°C for 5 min; (2) 39 cycles with three steps—denaturing at 94°C for 1 min, annealing at 55–67°C (depending on the specific primer sequence) for 1 min, and extending at 72°C for 2 min; (3) a final extension at 72°C for 5 min after the above 39 cycles. The amplified products were denatured at 95°C for 5 min, cooled on ice, and subsequently run on 8% denatured polyacrylamide gel at 70 W. One of accessions with a particular allele was randomly selected as the check to identify other accessions with the same allele. All checks and a molecular weight marker (PUC19 DNA digested by MspI) were included in the rest runs. Finally, the molecular weights of all the alleles were estimated by a gel-run including all detected alleles and a standard molecular weight marker of a 10 bp DNA Ladder (Invitrogen, Carlsbad, California). Gels were visualized by a silver-staining method. In the case of non-amplification, PCRs were repeated to exclude failures. If both PCRs failed, we recorded a null allele. The non-simple genotype could be classified into two types here: one with two alleles and one with more than two alleles. The proportion of the former was 7.61 ± 1.31%. The proportion of the latter was less than 1%. When more than two alleles occurred at a particular locus in one accession, they were amplified and run again. Two distinctly more stable and denser alleles in both experiments were selected for the structure analysis. If there were no distinctly more stable and denser alleles between two experiments, the genotype at that locus in the accession was recorded as missing.

Population structure and differentiation analysis

The model-based programs STRUCTURE (Pritchard et al. 2000; Falush et al. 2003) and INSTRUCT (Gao et al. 2007) were used to infer population structure using a burn-in length of 10,000, run length of 100,000. The former used a model allowing for admixture and correlated allele frequencies and the latter used a model allowing for population structure and population selfing rates. Ten independent simulations were run for each K (the number of clusters). The graphical display of the STRUCTURE results was generated using Distruct software (Rosenberg 2002; http://www.cmb.usc.edu/noahr/distruct.html). To determine the K value, we used both the LnP(D) value and Evanno’s ΔK (Evanno et al. 2005). LnP(D) is the log likelihood of the observed genotype distribution in K clusters and can be output by STRUCTURE simulation (Pritchard et al. 2000). The inferred k is ideal when the highest LnP(D) occurs, but it is sometimes difficult to find the highest LnP(D) before a much larger K is examined. In such cases, a clear knee is recommended, but it is usually somewhat subjective. Evanno’s ΔK took consideration of the variance of LnP(D) among repeated runs and usually can indicate the ideal K. The suggested Δk = M(|L(k + 1) − 2L(k) + L(k − 1)|)/S[L(k)], where L(k) represents the kth LnP(D), M is the mean of 10 runs, and S their standard deviation. Principal coordinate analysis was made using software NTSYS 2.1 (Rohlf 1997) to investigate the tridimensional structure of landrace rice in China. Phylogenetic reconstruction was based on the neighbor-joining method using Nei’s D A, an efficient distance estimator in obtaining correct topologies (Takezaki and Nei 1996), and implemented in PowerMarker version 3.25 (Liu and Muse 2004; http://www.powermarker.net), because the neighbor-joining method is especially suited for datasets comprising lineages with largely varying rates of evolution (Saitou and Nei 1987). Using inferred populations, a two level AMOVA (analysis of molecular variance) was performed in Arlequin ver 3.11 (Excoffier et al. 2005) with 1,000 permutations and sum of squared size differences as molecular distance. To investigate the population differentiation, F ST (Weir and Cockerham 1984) among populations was calculated and tested using FSTAT 2.9.3.2 (Goudet 2001).

Neutrality test

The Ewens–Watterson test of neutrality is able to detect deviations from a neutral equilibrium model. This is achieved by detecting whether there is either a deficit or an excess of expected homozygosity, estimated according to the observed allele frequencies (Obs.F) relative to that estimated according to null allele frequencies (Nul.F) by simulating random neutral samples (Ewens 1972; Watterson 1978). This test was performed using the program PopGene Version 1.31 (Yeh et al. 1999) and the algorithm given in Manly (1985). L95 and U95, which are, respectively, the lower and upper limits of the 95% confidence interval of Nul.F, were estimated through 1,000 simulations in this program. If Obs.F is within this confidence interval, the locus is neutral; otherwise, it is not.

Summary statistics

PowerMarker was also used to calculate the average number of alleles (N a), Nei’s gene diversity index (H sk) (Nei 1987), and polymorphism information content (PIC) (Anderson et al. 1993) values and stepwise mutation index. The average standardized molecular weight of the PCR products in each population was calculated as in Vigouroux et al. (2003). Allele richness per locus (R s , an estimator independent of the sample size; Hurlbert 1971) was also estimated. R s corresponds to the number of different alleles found when 2N gene copies are sampled at a particular locus. If 2N gene copies are totally examined at the locus, the expected number of different alleles in a sample size of 2n can be estimated with the formula:

$$R_{\text{s}} = \sum\limits_{i = 1} {\left[ {{{1 - \left[ \begin{aligned} & 2N - N_i \\ & 2n \\ \end{aligned} \right]} \mathord{\left/ {\vphantom {{1 - \left[ \begin{aligned} & 2N - N_i \\ & 2n \\ \end{aligned} \right]} {\left[ \begin{aligned} & 2N \\ & 2N \\ \end{aligned} \right]}}} \right. \kern-\nulldelimiterspace} {\left[ \begin{aligned} & 2N \\ & 2N \\ \end{aligned} \right]}}} \right]},$$

where N i represents the number of occurrences of the ith allele among the 2N sampled gene copies, and n is fixed as the number of individuals in the smallest group for the locus. The effective population sizes (N e) for populations at different altitudes and latitudes were estimated using LDNe (Waples and Do 2008).

Statistical analysis

The statistical significance of the difference of the estimators measuring the genetic diversity (including N a, R s, H sk, and PIC) and allele size was assessed using the t test and Wilcoxon paired test across loci. Using the inline applets of regression in Microsoft Excel, we also performed a regression analysis between genetic diversity (H sk and N a), effective population size (N e), and latitude/altitude. Using TFPGA (Miller 1997), we fulfilled the Mantel test (Sokal 1979) to estimate the correlation between genetic differentiations and geographic distances (among administrative districts), and between genetic differentiations and differences of latitude/altitude.

Results

Inference of genetic structure of cultivated rice in China

Among the 36 SSR loci we used to infer the genetic structure of cultivated rice, 543 alleles were totally detected in 3,024 varieties. The number of alleles per locus varied widely among marker loci, ranging from 2 (RM60) to 31 (RM247) with an average of 15.1. When we ran the STRUCTURE simulation using all 3,024 accessions, the LnP(D) value increased with K from 1 to 15, but showed an evident knee at k = 2 (Fig. 1a). This implied that there must be two distinctly divergent populations. There was a sharp peak of Evanno’s ΔK at K = 2, and a small peak at k = 6 (Fig. 1b–c). Due to the distinct divergence at K = 2 and the further increase of LnP(D), we performed further simulations using two inferred populations independently. Their ΔK indicated there were three subpopulations in each of the two inferred populations (Fig. 1e, g, and Fig. 2), corresponding to the six populations at k = 6 when simulation was run using all 3,024 accessions in one (Fig. S1). Simulations using INSTRUCT indicated that there was the lowest deviance information criterion at K = 2 and a secondary lower one at k = 6 (Fig. S2). The simulation using INSTRUCT and STRUCTURE gave consistent ancestry and structure patterns (Fig. S3). Thus, Chinese landrace rice firstly differentiated into two distinct populations, which correspond to the subspecies: indica and japonica (k = 2, Fig. 2). The individual membership between the predefined populations and the inferred populations (k = 3 within each subspecies in Fig. 2 and Fig. S4, and K from 3 to 6 in Fig. S1) indicated that three inferred populations in japonica corresponded to Ting’s soil-watery ecotypes, including lowland, upland, and medium (or intermediate) types; but those in indica corresponded to his seasonal ecotypes, including early, late, and medium (or intermediate) sown types (Fig. 2 and Fig. S1). Thus we defined the six inferred populations as six ecotypes: japonica lowland (Jap.L), japonica upland (Jap.U), japonica medium (Jap.M), indica early (Ind.E), indica late (Ind.L), and indica medium (Ind.M).

Fig. 1
figure 1

The average LnP(D) and ΔK over 10 repeats of STRUCTRUE simulations. a–c LnP(D) with k = 1–15, ΔK with k = 2–15, and ΔK with k = 3–15 for simulations using all 3,024 accessions; d–e LnP(D) with k = 1–15 and ΔK with k = 2–15 for inferred indica population; f–g LnP(D) with k = 1–15 and ΔK with k = 2–15 for inferred japonica population

Full size image
Fig. 2
figure 2

Relationship between the predefined and inferred populations. The two inferred clusters (k = 2) resulted from simulation using all 3,024 accessions in one and correspond to indica and japonica, respectively. Then three clusters (k = 3) were inferred within inferred indica and japonica independently. Naming convention of the predefined populations when k = 2: the leftmost symbols, I = indica, J = japonica; the symbols between two dashes, L lowland, U upland; the rightmost symbols: E early, M medium, L late. The labels when k = 3: JL japonica lowland, JU japonica upland, JN japonica unknown lowland or upland, IE early seasonal indica, IM medium seasonal indica, IL late seasonal indica, IN – indica unknowing seasonal type. Each vertical line represents one individual

Full size image

The principal coordinate analysis (Fig. S5) and neighbor-joining tree (Fig. S6) for 3,024 accessions both indicated that differentiation between two subspecies was the most apparent. Differentiation between soil-watery ecotypes within japonica was more apparent than within indica, and differentiation among seasonal ecotypes within indica was more apparent than within japonica. The more apparent differentiation among seasonal ecotypes within indica than within japonica was also confirmed by the differences in days to heading between three locations using 84 accessions (Table S2). This will be discussed later. More individuals of the predefined upland types in both predefined subspecies seemed to have admixed genomes; that is, among the predefined upland accessions, japonica contained a higher proportion of inferred indica genome (red circle at K = 2, Fig 2).

The continuous increase of LnP(D) after k = 3 within the inferred indica and japonica populations implied there are subtle sub-structures within the six ecotypes (Fig. 1d, f). Therefore, we performed further independent simulations within each of the six ecotypes. These revealed different numbers of geo-ecogroups within each of the six ecotypes. Investigation of these subtle differentiation within ecotypes will be a good subject for future research.

Genetic diversities and differentiation of inferred populations and their geographical distributions

The genetic diversity in each inferred population and subpopulation was measured by four estimators, viz. allele number (N a), allelic richness (R s), gene diversity (H sk), and PIC (Table 2). Japonica showed more variation in allelic richness (z = 1.69, P < 0.05), but lower heterogeneity as measured by H sk (z = 1.73, P < 0.05) than indica. Within japonica, lowland rice showed significantly lower H sk than upland rice (t = 2.044, P < 0.05); and within indica, early varieties showed significantly lower H sk than late varieties (t = 3.331, P < 0.01). The other estimators did not show significant differences between ecotypes within each subspecies. Genetic diversity was negatively correlated with latitude (N a, r 2 = 0.5696, P < 10−5; H sk, r 2 = 0.5554, P < 10−5) and altitude (N a, r 2 = 0.7325, P < 10−5; H sk, r 2 = 0.5714, P < 0.001) (Fig. 3). To eliminate the effect of subspecies (there were more japonica at high latitudes and altitudes than indica), we also examined these relationships within each subspecies and obtained similar results (Fig. S7).

Table 2 Genetic diversity of inferred ecotypes
Full size table
Fig. 3
figure 3

Regressions between genetic and geographic estimators. a Number of alleles and latitude, b Nei’s unbiased gene diversity and latitude, c number of alleles and altitude, d Nei’s unbiased gene diversity and altitude

Full size image

AMOVA conducted using inferred subspecies as groups, and inferred ecotypes within subspecies as populations indicated that 7.58% of the variation in landrace rice in China could be attributed to differentiation among ecotypes (Table 3), and 36.65% was attributed to the differentiation between indica and japonica. F ST among ecotypes showed that the differentiation among ecotypes was significant (Table 4). Both F ST (Table 4) and principal coordinate analysis (Fig S5) showed clearer differentiations among ecotypes within japonica than within indica. Mantel tests between genetic distances and geographical differences (measured by latitude (r = 0.3783, P < 0.01), altitude (r = 0.7323, P < 0.001), and geographical distance (r = 0.4423, P < 0.01)) exhibited significantly positive correlations (Fig. 4). Altitude had a large impact on genetic differentiation.

Table 3 Analysis of molecular variance (AMOVA) among inferred populations
Full size table
Table 4 F ST among six ecotypes
Full size table
Fig. 4
figure 4

Relationship between matrix of genetic distance and matrix of geographic estimators. a Nei’s genetic distance and difference in latitude, b Nei’s genetic distance and difference in altitude, c Nei’s genetic distance and geographic distance

Full size image

Neutrality test within different populations

Although SSR variation itself is mostly neutral, it has been reported that some SSR loci showed selective sweep reflecting selection or adaptation (Ellegren 2004). To detect whether there is selective sweep within each inferred population, we made neutrality tests using the program PopGene. Among the 16 loci that deviated from the null hypothesis of neutrality at P = 0.05, only four were shared by indica and japonica ecotypes; four were detected in one or two indica ecotype(s), and eight were detected in one or two japonica ecotype(s) (Table 5); more than half were detected in upland japonica, only three were detected in early indica.

Table 5 Neutrality test and the stepwise Mutation Index of 36 SSR loci
Full size table

Allele size in inferred populations

Directional evolution of microsatellites was reported in both animals (Rubinsztein et al. 1995) and plants (Vigouroux et al. 2003). We investigated the Stepwise Mutation Indices of the 36 loci (Table 5). Most of them showed evidence of stepwise mutation: 26 loci with SMI >0.80 (the threshold used by Garris et al. 2005), 18 loci with SMI >0.95, and 7 loci with SMI <0.7. Among 13 loci common to the study of Garris et al. (2005) and the present work, only rm219 and rm25 showed dramatically different stepwise mutation properties (Table 5). The average standardized allele sizes in japonica were significantly smaller (t = 39.18, P ≥ 0) than those in indica except for the two intermediate ecotypes which were not significantly different from each other (Fig. 5).

Fig. 5
figure 5

Average standardized molecular weights of microsatellites and standard errors for the six inferred ecotypes estimated using 26 SSR markers with a stepwise mutation index higher than 0.80

Full size image

Regression between allele size and altitude showed that allele size significantly decreased with increasing altitude (Fig. 6a). Similarly, allele size significantly decreased with increasing latitude, especially in the range 28°–34°N (Fig. 6b). To eliminate the effect of the distribution of subspecies, we studied the regression between allele size and altitude or latitude within each subspecies (Fig. S8). The correlations among latitudes were weaker than when using all accessions, and the correlation between allele size and altitude disappeared. In addition, the allele sizes in japonica were significantly smaller than those in indica at the same altitudes and latitudes (−0.34 vs. 0.46, t = 15.08, P < 10−30).

Fig. 6
figure 6

Regression between microsatellite allele size estimated using 26 SSR markers with SMI >0.80 and altitude (a) or latitude (b)

Full size image

Discussion

Genetic structure and reclassification of Oryza sativa L. in China

It is well known that cultivated rice has differentiated into two subspecies, indica and japonica (Oka 1988; Sano and Morishima 1992). In addition to the evident differentiation of subspecies, it is commonly accepted that there are other possible intra-specific differentiations. For example, Ting’s five-level taxonomic system (Ting 1957) described an hierarchical differentiation of cultivated rice in China. Although this system was commonly accepted in China and the differentiation between indica and japonica was proved at many levels, no molecular evidence has been found to support such intra-subspecific differentiation. Using varieties from parts of China and SSR markers, we earlier reported intra-subspecific differentiation patterns that differed from that of Ting (Zhang et al. 2007a, b).

Our current results confirmed the primary differentiation between indica and japonica within O. sativa. Structure simulations using STRUCTURE and INSTRUCT showed distinct structures at K = 2, at which the two subspecies were differentiated. The differentiation between the subspecies accounted for 36.6% of the genetic diversity in cultivated rice in China. Within the indica and japonica populations, however, the substructures were different from those indicated by Ting’s (1957) and Cheng–Wang’s (Cheng et al. 1984b) taxonomic systems, where seasonal ecotypes were firstly classified under both subspecies. In the present study, japonica appeared more distinct in terms of differentiation among soil-watery ecotypes (upland versus lowland), whereas indica was more clearly subdivided by seasonal ecotypes (early, medium versus late season). These different differentiations within the two subspecies may be attributed to their different growth environments and the corresponding cropping systems. It is well known that rice cultivars in temperate countries or regions like Japan, Korea, and northern China (roughly higher than 32°N) are exclusively japonica, whereas rice cultivars in tropical and subtropical regions are predominantly indica. In regions south of Yangtze River in China, both indica and japonica are grown, but indica is planted in the lowlands or valleys and japonica is grown on the hills (Xu et al. 1974; Khush 1997). For example, in Yunnan (21°8′32″ N–29°15′8″ N in southwest of China), indica varieties are mainly grown below 1,400 m, and japonica varieties mainly above 1,800 m (up to 2,200 m), with both types being grown between the two levels (Xu et al. 1974). This distribution led scientists to believe that indica is synonymous to tropical rice and japonica represents temperate rice.

At low altitude or low latitude where indica prevails, water resources are generally ample, thus it is not necessary to develop water-economical varieties, i.e., upland rice. Consequently, there is no apparent soil-watery differentiation within indica. In fact, there are quite fewer upland varieties in indica than in japonica under Ting’s taxonomic system. However, the bountiful heat and light resources available in many parts of south China provide conditions supporting the development of cropping systems with complicated seasonal patterns, such as double or triple cropping on an annual basis. Specifically adapted seasonal ecotypes are required to meet the needs of these intensive cropping systems. Such seasonal ecotypes derive their adaptation from different photothermic responses. The behavior of the various ecotypes are determined by three characteristics, viz the basic vegetative phase (BVP), photoperiod sensitivity (PS), and temperature sensitivity (TS) (Cheng et al. 1984a). Studies on BVP and PS were reviewed in detail by Vergara and Chang (1985). TS has been supported by only a few reports (AGRTPR 1978; Cheng et al. 1984a; Nakagawa et al. 2005). In general, shorter day-lengths and higher temperature can shorten days to heading. To estimate the photoperiod sensitivity and temperature sensitivity of the six inferred subgroups, we investigated the days to heading of 84 accessions sown at different locations (Sanya, Hangzhou, and Beijing; information on sowing date, day length, and temperature is shown in Table S2, and Fig. S9). The results showed that the days to heading clearly varied between the inferred subgroups in indica, but not in japonica (Table S2). The early varieties were strongly temperature-sensitive and weakly photoperiod-sensitive. From Sanya to Hangzhou, the days to heading were not distinctly changed due to the counterbalance of longer day-lengths and higher temperatures (Table S2). From Hangzhou to Beijing, the days to heading were distinctly prolonged due to longer day-lengths and lower temperature. The late varieties were strongly sensitive to both temperature and photoperiod, but the temperature sensitivity takes effect only when the day-length is short (especially shorter than 12 h). Thus the days to heading are distinctly prolonged from Sanya to Hangzhou, as day-lengths get longer and temperatures become lower. But difference of days to heading is less between Hangzhou and Beijing than between Sanya and Hangzhou, due to the less difference in day-lengths between Hangzhou and Beijing than between Sanya and Hangzhou, and the weaker temperature sensitivities of late varieties under long day-lengths (longer than 13 h, Fig. S9). From Sanya to Beijing, the obviously longer day-lengths (4 h) caused distinctly prolonged days to heading. Similar to the sensitivities to both day-length and temperature for the middle varieties, their differences of days to heading among three places lied between those for early varieties and late varieties. As noted by Vergara and Chang (1985), the days to heading for rice are also affected by other factors (such as seedling vigor), so days to heading within inferred subgroups show large variances. In summary, indica shows more differentiation among seasonal ecotypes than among soil-watery ecotypes.

In the current research, many varieties predefined as upland indica were identified as japonica through simulation of STRUCTURE; hence, more blue lines (red-circled) in the predefined populations I-U-E (indica, upland, early), I-U-M (indica, upland, medium), and I-U-L (indica, upland, late) than other predefined indica populations (K = 2, Fig. 2). This accords with the facts that varieties found in upland areas in Thailand, Burma, India, and other tropical Asian countries, usually show characteristics of japonica. Glaszmann (1987) found that most upland cultivars belonged to an enzymic group corresponding to japonica. Furthermore, a survey of native or traditional varieties of O. sativa in Africa by Kochko (1987) using isozymes (37 presumed loci) again indicated that varieties grown in upland field conditions tended to have genotypes like japonica. These reports demonstrate that the japonica type is more often associated with upland conditions, and implies that the origin of the japonica type might be related to adaptive selection under upland or non-standing water conditions, where limited resources of water, heat, and light required the development of varieties with drought tolerance. Other varieties planted in standing water evolved into lowland varieties. At the same time, insufficient light and heat restricted the development of diverse seasonal ecotypes within japonica. Nevertheless, because of the wide distribution from low to high latitudes and thus distinct differences in day length, japonica also showed weak differentiations for seasonal adaptation, but they are nearly accordant with geographical distribution. For example, japonica in south China mainly consists of late varieties, but northern varieties are predominantly early or medium types. In addition, the clearer differentiations among japonica ecotypes than among indica ecotypes, as revealed by F ST, principal coordinate analysis and selective SSR number, also indicate that japonica underwent more selection than indica during the domestication process. In the present study (Fig. 2), the majority of predefined upland japonica was inferred as upland type, whereas about half of the predefined lowland japonica was inferred as upland. This might reflect the different types of selection that were imposed on them. The upland type (in fields without standing water) is under directional selection in soil water regime, thus accumulating specific genotypes adapted to upland conditions. The lowland type (in fields with standing water) is not under directional selection, and thus includes genotypes that are both the same or different from upland genotypes. Thus varieties with lowland genotypes are seldom collected in upland fields, but varieties with upland genotypes do occur in lowland fields. In summary, japonica appeared more differentiations among soil-watery ecotypes than among seasonal ecotypes.

Our results therefore indicate a hierarchical differentiation and taxonomic system for cultivated rice (at least in China). The intra-specific differentiation of cultivated rice is as follows: indica and japonica were domesticated independently in tropical lowland areas and temperate upland areas, respectively. Complicated cropping systems and ample heat and light resources in tropical-like or lowland-like areas permitted the development of distinct seasonal ecotypes in indica, but in contrast, simple cropping systems and the restricted soil water regimes in temperate-like and/or upland-like areas drove the development of distinct soil-watery ecotypes with less clear seasonal ecotypes in japonica. This was consistent with other results using Chinese rice landraces from a part of China (Zhang et al. 2007a, b). Thus the hierarchical taxonomic system for cultivated rice should be subspecies (indica versus japonica), followed by ecotypes (distinct seasonal ecotypes for indica and distinct soil-watery ecotypes for japonica), and finally eco-geographical types. Further studies on the ecotypes and eco-geographical types may provide information on the utilization of intra-subspecific heterosis.

Directional evolution of microsatellites in cultivated rice in China

The phenomenon that new microsatellite mutations tend to increase molecular size has been documented in plants, such as chickpea (Cicer arietinum L.) by Udupa and Baum (2001), and maize by Vigouroux et al. (2003), and in humans (Weber and Wang 1993; Primmer et al. 1996; Cooper et al. 1999). In a separate study (Wang et al. 2008), the genetic structure of common wild rice (O. rufipogon Giff.) was analyzed using the same 36 loci as in the present study and 19 loci were proved to be stepwise (with a threshold of SMI = 0.80), Among those 19 loci, only Rm231 showed a clear difference in SMI (0.8704 for wild rice, 0.6392 for cultivated rice). Cultivated rice (O. sativa L.) appeared to have significantly higher molecular weights than its ancestor (the average standardized molecular weights were 0.29 and −0.99, respectively; t = 39.38, P ≥ 10−10). In the current study, cultivated rice had an evident further directional evolution in allele size in both subspecies and geography. The molecular weight of japonica was significantly smaller than indica, and also increased with decreased latitude and altitude. However, according to the hypothesis of directional evolution of microsatellite size, it is unclear whether directional evolution at SSR loci corresponds to the directional selection history of those populations, or, whether there are other interpretations for such results.

Do the populations with smaller allele size at SSR loci originate earlier than those with larger ones? Positive answers were given by others (Udupa and Baum 2001; Vigouroux et al. 2003; Weber and Wang 1993; Primmer et al. 1996; Cooper et al. 1999). However, for populations within cultivated rice, such as subspecies and other populations, the reasons might be different. Common wild rice was perennial with both sexual and asexual reproductive strategies; the asexual role was dominant under natural conditions (Oka 1992; Xie et al. 2001). Thus wild rice had fewer generations to evolve than its cultivated counterpart over the same time periods. If microsatellite mutations showed an almost 2:1 ratio of gain in size over losses (Banchs et al. 1994; Weber and Wang 1993), SSR alleles should be smaller in the later originating populations of cultivated rice than in those originating earlier, assuming they originated from the same or similar populations of common wild rice and that domestication events occurred in the time longer than several thousands years. If so, indica (with larger alleles) should have originated earlier than japonica in China, as was proposed by Ting (1957); and indica should be more similar to wild rice in many aspects such as morphology, genetic material, and growth habits. It is well known that the distribution of the two subspecies is related to altitude and latitude and allele size is correlated with altitude and latitude (Fig. 6, Fig. S8). Hence we need to determine how much of the difference in allele size between the two subspecies is related to their different occurring times of domestication and how much it relates to changes after domestication. The most influential environmental effects probably relate to geography (altitude and latitude), under which adaptation to the specific natural conditions occurred. Comparison of the different allele sizes between the two subspecies with those at the same altitudes and latitudes indicated that the environmental difference accounted for about 30% of the difference in allele size between the two subspecies: {[0.55 − (−0.61)] − [0.46 − (−0.34)]}/[0.55 − (−0.61)]. This was also indicated by the regression of allele size and altitude or latitude using all 3,024 collectively, or using each subspecies separately (Fig. 6, Fig. S7). More evidence, especially from the wild rice, is needed to prove whether the other 70% of the difference in allele size between the two subspecies derives from their different times of domestication.

Temperature is the most obvious environmental factor that changed over different latitudes and altitudes. The general macroevolutionary patterns show that most higher taxonomic groups originate in the tropics where temperatures are high (Jablonski 1993), and that speciation rates decrease with decreasing temperatures from the equator to the poles (Flessa and Jablonski 1996; Stehli et al. 1969). Higher temperature is expected to result in higher mutation rate (Bleiweiss 1998; Gillooly 2005). We observed a negative correlation between allele number and altitude/latitude (Fig. 3a, c and Fig. S7). It is possible that populations at lower altitude/latitude subjected to higher temperature exhibit higher mutation rates, and consequently have larger SSR allele sizes. A negative correlation between allele size and altitude was also found in maize (Vigouroux et al. 2003).

In addition, we speculate that selection was a contributing factor to the directional evolution of some SSR loci. Under more rigid conditions at higher altitude/latitude, newly occurring variations are seldom inherited to the next generation due to their rare frequency, and thus increase in the gain in allele size is slower than at lower altitudes/latitudes. In our current study, the number of alleles and Nei’s diversity index at low altitudes/latitudes were higher than those at high altitudes/latitudes. Selection or founder effects are two main ones among the possible reasons for these phenomena. At high latitudes, the fewer variations (Fig. S7) and smaller effective population size (Fig. S10) might imply that the decrease in genetic diversity with latitude mainly results from the founder effect occurring during dispersal of cultivated rice from low latitude to high latitudes (The dispersal has been argued by researchers, see review by Khush 1997). However, at higher altitudes, the fewer variations but larger effective population sizes (Fig. S10) imply that SSR variations suffer from more intense selection under more rigid conditions, such as lower temperature, less heat, and lower water availability. Given that most SSR variations are neutral, the argument for selectivity of SSR may make some researchers uncomfortable, but it has been reported that some SSR loci could reflect selection or adaptation effects (Ellegren 2004). Tam et al. (2005) reported that 18.75% of SSR loci are non-neutral in a tomato collection whereas all SSR loci were neutral in a pepper collection. In the current work, the proportion of non-neutral SSR loci was 44% across all ecotypes, and ranged from 8 to 25% within ecotypes. Seventeen of the SSR markers were also used by Gao and Innan (2008). Among 10 non-neutral loci, 2 (rm247 and rm258) were selective in both studies, but 7 additional loci were detected only by us, and 1 by Gao and Innan. The different neutralities for the same loci detected in different studies could result from different selective events occurring in different populations. Of course, it may be too early to conclude a “selective sweep” for a specific locus if we simply detect a deviation from neutrality.