Introduction

The development of rice production in the dry winter season in Bangladesh, locally known as boro, using groundwater for irrigation coupled with Green Revolution high yielding rice varieties (HYVs) and use of chemical fertilizers played a key role in tripling the annual rice output in Bangladesh from about 9 million tons in 1970–71 to more than 25 million tons by the mid-1990’s (BBS 2004), making Bangladesh self-sufficient in this staple food grain (Baffes and Gautam 1996). Today, boro season rice contributes about 50% of the total annual rice production (MOA 2008). Recent findings have shown elevated As concentrations in rice grain from many parts of Bangladesh (Lu et al. 2009; Meharg and Rahman 2003; Williams et al. 2006; Zavala and Duxbury 2008), indicating that rice contributes to human exposure to As, in addition to the well known problem of As ingestion through drinking water (BGS/DPHE 2000). The average daily consumption of rice by a Bangladeshi male adult is high, 453 g milled rice per person per day (Bangladesh Agricultural Research Council, unpublished food and nutrition data, 2007), providing more than 70% of daily calorie intake (Ninno and Dorosh 2001). Thus, contaminated rice may represent a significant pathway of As consumption by humans, in addition to drinking water (Duxbury et al. 2003; Meharg and Rahman 2003).

Large variations in grain As concentration of Bangladeshi rice varieties sampled from rice fields have been observed (Duxbury et al. 2003; Zavala and Duxbury 2008). Market basket surveys also revealed similar variability (Williams et al. 2005, 2006). Some of this variability has been explained by differences in groundwater irrigation levels of As (Williams et al. 2006) and baseline soil As concentrations (Lu et al. 2009). A recent study indicated that variation in rice grain As concentration in Bangladesh was largely controlled by rice genetics (Norton et al. 2009b), but expansion of this work to India and China (Norton et al. 2009a) led to the conclusion that that environment played a more important role than genetics. However, quantitative characterization of the relative importance of rice genetics (G), growing environments (E) and GxE interactions (GEI) on rice grain As concentrations has not been done. The characterization of E is particularly important because As uptake by rice mainly depends on As availability rather than total As in the soil (Sheppard 1992). Rice grain can accumulate relatively large amounts of As even from soils not contaminated by As (Daum et al. 2001). Soil factors that have been reported to regulate As bio-availability to plants include: pH and redox status (Marin et al. 1993), clay content (Sheppard 1992), existence of poorly crystalline iron oxide (Bogdan and Schenk 2009; Takahashi et al. 2004), and type and amount of organic matter in soil (Mitchell and Barre 1995). These soil properties vary both at both local and regional scales in rice growing areas of Bangladesh (Brammer 2009).

The objectives of the present study were to (1) genetically classify and characterize diversity in rice varieties released by the Bangladesh Rice Research Institute (BRRI), (2) to understand the role of G, E and GEI with respect to As concentration in rice grain of BRRI varieties grown in diverse soil environments in the two main rice growing seasons and (3) to identify rice varieties with low grain As concentration suited to specific or broad growing environments in Bangladesh.

Material and methods

Rice varieties and genetic classification

Twenty dry winter season (locally known as boro) and 18 wet monsoon season (locally known as aman) rice genotypes developed by the Bangladesh Rice Research Institute (BRRI) were used in this study (Table 1). These HYVs have been developed for specific environments and cropping systems in the boro (January to May) or aman (July to December) seasons. All 38 test varieties were evaluated with 30 simple-sequence repeat (SSR) markers and genetically classified to subpopulation based on clustering with a control set of 50 rice accessions whose sub-population identity was known (10 each of the temperate japonica, tropical japonica, indica, aus and basmati/aromatic (Garris et al. 2005)) using a model based Bayesian approach (Falush et al. 2003; Pritchard et al. 2000) and the C.S. Chord distance method (Cavalli-Sforza and Edwards 1967).

Table 1 Rice varieties used
Full size table

Experimental sites

Rice grain samples were collected in 2004–2005 from rice varietal evaluation trials at nine BRRI research stations located at Barisal, Bhanga, Comilla, Gazipur, Habiganj, Rajshahi, Rangpur, Satkhira and Sonagazi, and at the Bangladesh Agricultural Development Corporation station in Kustia. The sites represent nine agro-ecological zones and the major rice growing environments in Bangladesh (Fig. 1). Boro (20 genotypes) and aman (18 genotypes) rice varieties were grown independently in their respective seasons at all locations, except for BR3 that was grown in both seasons. Rice was not grown at Bhanga and Habiganj in the aman season due to deep flooding nor at Sonagazi in the boro season due to high soil salinity in this dry season. Poor seed germination prevented the growth of BR27 at Bhanga and Kustia, BR6 and BR27 at Rajshahi and BR9, BR18 and BR19 at Satkhira. In both seasons, the fields were flood-irrigated to maintain 5–10 cm standing water throughout the growing season until water was drained one week before harvest. Groundwater from shallow tubewells (STW’s; <100 m) containing 120 and 234 μg As L−1 was used for irrigation at Bhanga and Satkhira, respectively. Irrigation was from a surface water canal at Barisal and from deep tube wells (>100 m) at other sites with As concentrations <10 μg L−1 (Table 2).

Fig. 1
figure 1

Location of the ten experimental sites at Bangladesh Rice Research Institute stations

Full size image
Table 2 Soil and irrigation water properties of experimental sites
Full size table

Experimental design and management

At each site, the experiment was laid out in a randomized complete block design with three replications and 5 × 5 m plots. Fertilizers were applied at elemental rates of 120-30-60-10-2 kg ha−1 of N-P-K-S-Zn and 100-30-60 kg ha−1 of N-P-K for boro and aman crops, respectively, using urea, triple super-phosphate, muriate of potash, gypsum and zinc sulfate. Nitrogen was applied in three equal splits, a basal dose before transplanting, at active tillering and at panicle initiation. The entire amounts of P, K, S and Zn fertilizers were applied as basal doses. Rice seedlings, 40–50 days old in boro and 30–40 days old in aman, were transplanted into puddled soil at the rate of 2–3 seedlings per hill with a hill-to-hill spacing of 20 × 20 cm.

Rice yield and grain arsenic analysis

At each location, mature rice was harvested from a 5.2 m2 area at the center of each plot. Grain weight data were adjusted to 14% moisture level and rice yield computed as yield per hectare of rough rice. Grain samples for As analysis were taken randomly from panicles of the mature rice plants within the harvest area. After cleaning, the grain samples were oven dried at 70°C then husked and polished by a mini rice mill to white rice. Grain samples were digested with a concentrated HNO3-30% H2O2 mixture (Tang and Miller 1991) and analyzed for total As by inductively coupled plasma mass spectrometry (ICP-MS; Agilent Technologies model 7500, USA). Replicated reagent blanks and a National Institute of Standards and Technology (NIST) certified reference rice flour (SRM-1568a) were included in each analytical batch for quality control. The range of recovery of total As from SRM-1568a was 91%–106% (Online Resource Table 2).

Soil sampling and analysis

At each site during harvest, composite soil samples (0–15 cm soil depth) were collected from five points (four from two diagonals and one from the intersection point of the diagonals) within the harvest area. The soil samples were mixed, air-dried, and passed through a 2 mm sieve prior to chemical analysis. Soil pH was measured using a glass electrode and a soil:water ratio of 1:2.5. Soil organic C (SOC) was determined by dry combustion using a Europa Scientific Roboprep C and N analyzer. Soil Fe parameters measured were total Fe (TFe) by HNO3-H2O2 digestion (USEPA 1990), poorly crystalline Fe-oxide (AOFe) by extraction with acidified ammonium-oxalate in the dark and total Fe oxide (CDFe) by extraction with citrate-dithionite (Scheinost and Schwertmann 1999). Soil As parameters measured were total As (TAs) by HNO3-H2O2 digestion (USEPA 1990), ligand-exchangeable As (PAs) by Na- phosphate extraction and As associated with poorly crystalline Fe oxides (AOAs) by oxalate extraction (Loeppert et al. 2002). Total As in soil digests and extracts was determined by ICP-MS and Fe was measured by ICP-atomic emission spectroscopy (Spectro-CCD, USA). The recovery of total As from NIST certified reference soil SRM-2709 ranged from 99%–100%.

Statistical analysis

To examine the effect of environment (E) and genotype (G) on rice grain-As concentration, a combined analysis of variance (ANOVA) was performed using the general linear model (GLM). Locations and genotypes were considered as fixed effects, while replications were random effects. The sum of squares was first partitioned into genotype (G), environment (E) and genotype-environment interaction (GEI). Then, the Additive Main Effects and Multiplicative Interaction (AMMI) model, that combines standard ANOVA for G and E with principal component (PC) analysis, was used to decompose the GEI into several interaction principal analysis axes (IPCAs) (Crossa et al. 1990; Gauch 1992.) The relative contribution (expressed in percentage) of each component to the total variation was computed. MATMODEL software Version 3.0 (Gauch 2007) was used for ANOVA and AMMI analysis. A PC analysis was also done on the correlation matrix of soil variables. Correlation analyses were conducted to test for linear relationships between the soil co-variables and the E mean of grain As.

Results

Genetic classification of rice varieties

Thirty two (90%) of the BRRI varieties used in this work were classified into the indica subpopulation, two (BR2 and BRRI dhan 33) were identified as admixtures between indica and aus, and four were classified into the genetically divergent aromatic/Group V subpopulation (Fig. 2). The Group V types were all aman season varieties. The indica types clustered into two distinct groups, identified as indica-1 and indica-2 in the unrooted tree in Fig. 2. The average number of alleles/locus detected among the 34 indica/aus cultivars was 3.6, the average gene diversity (He) was 0.33 and the average polymorphism information content (PIC) was 0.29. Previously published estimates of diversity based on evaluation of 79 indica varieties by Garris et al. (2005) showed an average of 7.3 alleles/locus, average He of 0.55 and average PIC value of 0.52. Similarly, the average number of alleles/locus detected among the 4 aromatic/Group V cultivars was 1.4, the average He was 0.15 and the average polymorphism information content (PIC) was 0.11. Previously published estimates of diversity based on evaluation of 19 aromatic/Group V varieties by Garris et al. (2005) showed an average of 3.4 alleles/locus, average He of 0.39 and average PIC value of 0.38. These results indicate that there is a great deal of genetic variation available within the indica and aromatic/Group V gene pools that is not represented within the BRRI varieties.

Fig. 2
figure 2

Unrooted neighbour-joining tree showing the genetic relationship among the investigated rice genotypes, based on CS Chord (Cavalli-Sforza and Edwards 1967) distance calculated using thirty nuclear microsatellite loci. Admixed individuals are identified with a black line. The scale indicates genetic distance

Full size image

Soil properties

Many of the physical and chemical properties of the soils varied considerably (Table 2). Soil pH ranged from 5.0 at Habiganj to 7.5 at Satkhira. Soil total As ranged from 4.4 mg kg−1 at Rangpur to 19.6 mg kg−1 at Bhanga where high As groundwater was used for irrigation. Arsenic extracted by Na-phosphate ranged from 0.5 to 3.4 mg kg−1 and averaged only 11% of total As. The highest fraction of total As extracted by Na-phosphate (23%) was for the Satkhira soil. The concentrations of poorly crystalline Fe oxide ranged from 0.18% at Rangpur to 1.66% at Habiganj. The amount of As associated with poorly crystalline Fe oxide ranged from 0.9 to 8.7 mg kg−1 (20%–56% of total As) with the highest amount at Bhanga and Satkhira,. The ratio of Fe:As extracted by ammonium oxalate varied six-fold from 3168 at Habiganj to 527 at Satkhira.

Principal component analysis (PCA) of the soil properties identified 2 principal components with eigenvalue >1, and these components explained 79% of the total variance. The biplot of the first 2 PC-scores shows clear separation of experimental sites (Fig. 3), which also shows the vector loading of soil variables. Only one grouping of sites (Gazipur, Rajshahi and Rangpur) was observed. The soil variables that separated these three sites from other sites (along the PC1 axis; explaining 48% of total variance) were soil texture, SOC and Fe and As related soil properties. Soils at these sites have been altered by addition of river bottom sediments to create “new” surface soil. The separation of Satkhira from Habiganj along the PC2 axis (explaining 31% of total variance) was largely influenced by soil pH, soil As and Fe-chemistry.

Fig. 3
figure 3

Biplot of the first two principle components scores of soil properties and weighted loading among the soil variables at 10 experimental sites

Full size image

Variation in rice grain As concentration

In the aman-season, the mean grain As concentration of rice varieties grown at different sites ranged from 0.10 mg kg−1 for the indica variety BR 23 to 0.22 mg kg−1 for the admixed variety BRRI dhan 33 (Table 3). The strongly photosensitive varieties (Table 1), all four of the aromatic/Group V varieties (BR5 and BRRI dhan 34, 37 and 38) and five of the indica varieties (BR 4, 22, 23 and BRRI dhan 40 and 41) had the lowest grain-As concentrations, while the four indica varieties with low photosensitivity (BR 25 and BRRI dhan 31, 32 and 39) and the admixed variety BRRI dhan 33 with moderate photosensitivity had the highest grain As concentrations. The mean grain As concentration in the boro season varieties ranged from 0.22 mg kg−1 for BR 3 to 0.34 mg kg−1 for BRRI dhan 35 (Table 4), and the overall mean (0.29 mg kg−1) was 87% higher (F-value = 503, p < 0.0001; Online Resource Table 3) than that (0.15 mg kg−1) in the aman season. Grain As concentration in BR 3, the one genotype that was grown in both seasons, was also significantly higher (F-valueseason = 185, p < 0.001) in the boro-season than in the aman-season (Fig. 4).

Table 3 Mean grain As concentration (mg kg−1) of 18 rice varieties grown in 8 environments in the aman season
Full size table
Table 4 Mean grain As concentration (mg kg−1) of 20 rice varieties grown in nine environments in the boro season
Full size table
Fig. 4
figure 4

Mean grain As concentration of BR3 at different sites in the two growing seasons. Error bar represents the standard deviation of the mean. BAR = Barisal, BHN = Bhanga, COM = Comilla, GAZ = Gazipur, KUS = Kustia, RAJ = Rajshahi, RAN = Rangpur, and SAT = Satkhira

Full size image

Effect of genotype (G), environment (E), and GE interaction on grain As concentration

Analysis of variance showed that the As concentration in rice grain was significantly affected by E, G and GEI (Table 5). Environment explained 80 and 69% of the variation in grain-As concentration in aman and boro seasons, respectively. In contrast, G explained only 9 and 10%, and GEI 10% and 21% of the variability in grain-As concentration in the two seasons. Proportioning of GEI using the AMMI model showed that the first four interaction principal component analysis axes (IPCA) were highly significant (p < 0.001) in both seasons, with the first two IPCA capturing the major portion (70%–80%) of the variation (Table 5).

Table 5 AMMIa analysis of variance for the significance of the effects of genotype (G), environment (E), and genotype x environment interaction (GEI) on grain As concentration for 18 aman and 20 boro season genotypes, and partitioning of GEI into AMMI IPC axes
Full size table

The AMMI biplots of IPCA1 versus mean grain As concentration for genotype or environment facilitate interpretation of the effects of E, G and GEI (Fig. 5a and b). In the bi-plot, genotypes or environments with similar IPCA1 scores have similar GxE interaction patterns. Rice genotypes with IPCA scores >0 have increased and decreased grain-As concentrations in environments that have IPCA1 scores >0 and <0, respectively. The reverse applies for genotypes with IPCA1 scores <0. Genotypes with IPCA1 scores close to zero have the least variation in grain As concentration across environments, i.e. they are the most stable across environments, regardless of their As concentration levels.

Fig. 5
figure 5

AMMI biplot of mean grain As concentration and the first principal component axis (IPCA 1) scores of (a) 18 aman season genotypes and 8 environments (sites) and (b) 20 boro season genotypes and 9 sites. The vertical dashed lines represent the grand means over all sites. Sites are Barisal (BAR), Bhanga (BHN), Comilla (COM), Gazipur (GAZ), Habiganj (HAB), Kustia (KUS), Rajshahi (RAJ), Rangpur (RAN), Satkhira (SAT), and Sonagazi (SON). The rice variety numbers correspond to Table 1

Full size image

In the aman season, the late-flowering, photosensitive varieties (BR 4, 5, 22, 23 and BRRI dhan 34, 37, 38, 40 and 41) (BRRI 2004) had mean grain As concentrations lower than the overall mean and interacted differently with environment in terms of As accumulation in grain (Fig. 5a). Of this low grain-As group, the varieties BR 4, 23 and BRRI dhan 38 showed the greatest stability across environments and therefore could be considered the best varieties for limiting the amount of grain-As during the aman season. The variety BRRI dhan 33, which had the highest grain-As concentration and a large positive IPCA1 score, would be the poorest choice for the aman season (Fig. 5a). In the boro season, BR3 would be the best varietal choice because it had the lowest mean grain As-concentration and showed high stability across environments, while BRRI dhan 27 and 35, with high grain As concentration and fairly high positive and negative GEI, respectively, would be the poorest choices (Fig. 5b).When BR3 was grown in the aman season it had a mean grain As concentration close to the overall mean and had a high interaction with environment (Fig. 5a)

Satkhira in the aman season and Satkhira and Habiganj in the boro season (Fig. 5a and b) were clearly separated from other sites. Grain As concentrations were generally the highest at Satkhira in both seasons (Tables 3 and 4), with mean values of 0.454 and 0.362 mg kg−1 compared to the overall environment means of 0.288 and 0.154 mg kg−1 in the boro and aman seasons, respectively. Grain As concentrations in the boro season were the lowest at Habiganj, with a site mean of 0.157 mg kg−1.

The Satkhira site had very large negative IPCA1 scores in both aman and boro seasons (Fig. 5a and b) and the greatest interaction with genotypes. Rice varieties BR25, and BRRI dhan 32 and 39 in the aman season, and BR 12, 14, 15 and BRRI dhan 27 in the boro season accumulated relatively higher amounts of As in their grain at Satkhira compared to the other sites. In contrast, varieties BR 3 and BRRI dhan 40 and 41 in the aman season and BR 1, 2, 7 and 26 in the boro season had positive IPCA1 scores and showed the least relative increase in grain As at Satkhira compared to other sites. With the exception of Satkhira, the site means of grain-As were more tightly clustered in the aman season (range from 0.10 to 0.15 mg kg−1) than in the boro season (range from 0.24 to 0.36 mg kg−1 for common sites). Only Gazipur and Rajshahi in the boro season had IPCA1 scores close to zero, indicating little interaction with genotypes.

Correlation of grain-As with soil variables

The environmental (site) means of grain-As concentration were positively correlated with soil PAs in both seasons (r = 0.840 and 0.806; p < 0.01), with AOAs in the aman season (r = 0.815, p < 0.05) and with soil pH in the boro season (r = 0.811; p < 0.01); and negatively correlated with the ratio of AOFe to AOAs in the boro season (r = −0.843, p < 0.01; Table 6). When the site means of grain-As concentration were regressed on the scores of the first two PCs of soil variables, only PC2 scores showed a significant linear relationship, which explained 82 and 48% of the total variability in grain-As concentration in the boro and aman seasons, respectively (Table 7). The soil variables with loading >±0.3 to PC2 were pH, silt content, AOFe, PAs, and the ratio of AOFe to AOAs, respectively (see Online Resource Table 4). PC2 was also correlated with IPCA 1 (r = −0.669, p < 0.05), indicating that these same soil variables affect the GEI.

Table 6 Correlation coefficients (r) between environmental mean (E-mean) of grain As and soil properties
Full size table
Table 7 Results of linear regression analysis of environmental mean of grain As versus PC1 and PC2 scores of soils variables
Full size table

Discussion

Rice varieties released by BRRI belong primarily to the indica subpopulation, with a few aromatic/Group V cultivars. SSR marker analysis showed that the indica varieties grown in the aman and the boro seasons are genetically very closely related (Fig. 2), while the aromatic/Group V varieties are genetically distinct and differ for traits such as flowering time and photo-period sensitivity, which are important components of their adaptation to the aman growing season. Late-flowering varieties in aman season, whether belonging to the aromatic/Group V or indica subpopulations, had significantly lower grain As levels (F-value = 6.83, p < 0.0001; Online Resource Table 3) than varieties that flowered early, but the reasons for this are unclear at the present time.

When BRRI rice varieties were grown over ten different environments in Bangladesh, most (69%–80%) of the variation in grain As concentrations was explained by the environment. This result appears to be similar to that reported by Norton et al. (2009b) for 13 rice varieties grown at six sites in three different countries (Bangladesh, India and China), although Norton et al. did not quantitatively partition the variability caused by G, E and GEI. In a separate study using 76 cultivars evaluated at two locations during the boro season in Bangladesh, Norton et al. (2009a) observed strong genetic control over grain-As. A weakness of this conclusion is that two sites are insufficient to capture the impact of environmental variation on rice grain arsenic levels, even though genetically more diverse materials were evaluated than in the present study. Indeed, notable differences in ranking of BRRI varieties occurred between the two studies (Online Resource Table 5); for example, our high rankings (low grain-As) of BR 6 and BRRI dhan 28 and low rankings of BR 8 and BRRI dhan 35 were reversed in the Norton et al. (2009a) study. We note, however, that the highest and lowest levels of arsenic found by Norton et al. were detected in local aus land races and japonica materials, respectively, and neither of these sub-populations was well represented in the present study. Further, our data is for polished rice. Polishing could influence differences in grain-arsenic between varieties but is relevant as most rice consumed is polished or white rice. Overall, we conclude that both studies are too limited to conclusively establish the extent to which there are opportunities for reducing arsenic in rice grain through rice genetics. Nevertheless, our results can be applied to BRRI rice varieties which are, of course, commonly grown in Bangladesh.

Although the general observation from the present study was that varietal grain-As varied greatly across environments, there were a few varieties that performed consistently across environments. Thus, some general recommendations can be made about which rice varieties to grow and which to avoid from a grain As perspective. In the boro season, BR3, a high-yielding variety released in 1973, had the lowest overall mean grain As concentration in both our study and that of Norton et al. (2009a), and was the most stable across environments. This variety can be recommended as a low grain arsenic variety for the boro season. BR3 was formerly a popular variety, being cultivated on 19% of the boro production area in 1990 (Hossain et al. 2006). However, its use has declined substantially since the release of BRRI dahn 28 and 29 in 1994, both of which have high yield, good grain quality and shorter maturity. In 2000, BRRI dahn 28 and 29 occupied 11 and 9%, respectively, of the boro area (Hossain et al. 2006), and their extent has likely increased since then. Changing from BRRI dahn 28 and 29 to BR3 would reduce grain arsenic by an average of 21%–26%. Unfortunately, other studies have shown that both BR 3 and BRRI dhan 29 are susceptible to As toxicity, while BRRI dhan 28 is less vulnerable (Ahmed 2009; Biswas 2009). The variety BRRI dhan 47, which had relatively low grain As concentrations in the study by Norton et al. (2009a) and in a pot study with soils from a farmer’s field that varied in arsenic levels (Biswas 2009), may also prove to be a good choice for farmers, as it is tolerant of high soil As (Panaullah and Duxbury, personal communication).

The most popular BRRI variety in the aman season is BR11, which has the highest yield potential (6.5 Mg ha−1; Online Resource Table 1) of all of the BRRI aman season varieties. This variety was released in 1980 and occupied 42% of the rice area in 2000. In our study, it had an average grain As concentration of 0.164 mg kg−1 and a mean site weighted ranking of 12th out of the 18 varieties grown. The variety BR 23, had the lowest grain As concentration at 0.101 mg kg−1, and is a strongly photo-period sensitive variety with salinity tolerance that is also recommended for late planting (typically caused by flooding). It is, however, somewhat vulnerable to As toxicity (Panaullah and Duxbury, personal communication). BRRI dhan 41, also a salt tolerant variety, had the second lowest mean grain As, 0.117 mg kg−1, and the highest mean grain yield (4.1 Mg ha−1) in the study year, so it would be a good choice for the aman season, provided that it is not overly vulnerable to As toxicity. Unfortunately, all of the high-yielding, short duration rice varieties which were bred to promote crop intensification (BRRI dhan 32, 33 and 39) had relatively high mean grain As concentrations (0.17–0.22 mg kg−1).

Making varietal recommendations is complicated by farmers preferences for agronomic performance and grain yield potential, and by consumer preferences for rice grain quality characteristics. For these reasons, not all BRRI varieties are widely grown in Bangladesh, and varieties demonstrated to accumulate low levels of grain-As may not be popular or commercially viable. For varieties which performed differently across environments, extensive evaluation across years and environments would be needed to guide choices for specific locations. Moreover, given the narrow genetic base of current BRRI varieties, a plant breeding program to reduce rice grain As should explore a wider gene pool than has currently been exploited in the BRRI breeding program.

Mechanistic explanations for G, E and GEI effects on As concentration in rice grain would greatly aid development of sound strategies for reducing human exposure to As. The dominance of environment or site in controlling grain As concentration in BRRI varieties suggests that As levels and speciation in soil solution coupled with soil solution chemistry are important factors controlling As uptake by roots and ultimately levels in rice grain. Poorly crystalline Fe oxide (AOFe) concentration, total As and plant available P were soil variables contributing to a multiple regression prediction of As concentrations in rice grain and straw (Bogdan and Schenk 2009). We also found that measures of AOFe and soil As were correlated with the environmental mean of grain As concentration (Table 6), although in our case significant relationships were found with phosphate extractable As (both seasons), AOAs (aman season) and the ratio of AOFe/AOAs (the ratio of poorly crystalline Fe oxide to As associated with this fraction) in the boro season. Soil pH, which has a strong effect on adsorption of arsenate but only a small effect on adsorption of arsenite on Fe oxides (Dixit and Hering 2003), was also correlated with grain As in the boro season. A possible explanation for the relevance of the AOFe/AOAs ratio is that incomplete reduction of amorphous Fe oxides (van Bodegom et al. 2003) would leave residual adsorption capacity for inorganic As species and phosphate. The capacity for readsorption of solubilized As could reasonably be expected to increase with increasing levels of AOFe and a widening ratio of AOFe/AOAs.

The extent of formation of Fe-oxide precipitates on rice roots and in rhizosphere soil may also be an important factor as this leads to adsorption of As from soil solution (Hossain et al. 2009; Liu et al. 2004, 2006) and reduces As uptake by rice (Mei et al. 2009; Ultra et al. 2009). Because rice varieties differ in their release of oxygen to the rhizosphere (Mei et al. 2009), there could be a genetic interaction with Fe availability across sites that contributes to GEI. Arsenate, rather than arsenite, is the main species adsorbed on Fe-oxide root precipitates (Liu et al. 2006), which is a positive factor as arsenate adsorption will be greater than that of arsenite in the acidic pH environment of the rice rhizosphere (Begg et al. 1994; Dixit and Hering 2003).

Several soil chemical factors that could affect As uptake by rice, especially phosphate and silicate, were not measured in the present study. Phosphate competes with arsenate (Meharg and Macnair 1992) and silicate competes with arsenite (Ma et al. 2008) for uptake by roots. Phosphate also competes with inorganic As species for adsorption sites (Wauchope and McDowell 1984) and its effects on As uptake can be either positive or negative, presumably depending on the levels and species of each in soil solution. Concentrations of As in rice were found to decrease with increasing soil solution Si concentration (Bogdan and Schenk 2008) and silicon fertilization (Li et al. 2009) in studies where soil Si availability appeared to be low. Temporal variability in inorganic As speciation in rice fields has been observed over the growing season with arsenate being the main form initially but with arsenite dominating as soils are reduced (Panaullah et al. 2009; van Geen et al. 2006), so the importance of competition with phosphate and silicate will also be temporally dynamic. Rice plants show genetic differences in Si uptake depending on the expression levels of the Lsi1 and Lsi2 transporter genes (Ma et al. 2007), which could also lead to genetic differences in As uptake. Understanding how these genes are expressed under different environmental conditions could also be important to limiting the uptake of As by rice.

As in previous studies (Duxbury et al. 2003; Williams et al. 2005), we found that grain As was higher in the boro than the aman season; in our case the overall mean was almost twice as high (0.288 versus 0.154 mg kg−1). This same pattern was found with BR 3, the only variety grown in both seasons (Fig. 4). It has been suggested that the seasonal differences in grain As concentration could be caused by irrigation of boro rice with As contaminated groundwater (Duxbury et al. 2003; Williams et al. 2006), however this explanation does not fully explain the differences observed in the present study because irrigation water contained negligible levels of As at five of the seven sites where this was measured (Table 2). Inputs of As through irrigation water may have contributed directly, or indirectly through build up of As in soil, to high grain As at the Satkhira and Bhanga sites where irrigation water contained 120 and 234 μg As L−1, respectively. Lower levels of grain-As in varieties grown during the aman season could also be due to a dilution of soil solution-As by As-free rainwater. Interestingly, BR3 which had the lowest mean grain-As concentration and the greatest stability across environments in the boro season, had a mean grain-As concentration close to the overall mean and was highly variable across sites in the aman season. This result suggests that (i) there are important genetic differences between BR3 and aman season varieties that have lower grain-As than BR3, and (ii) interactions between genetics and non-soil environmental factors, such as temperature and light, affect As uptake and/or utilization by rice.

The myriad of soil chemical and genetic factors that could contribute to As solubility in soils, its uptake by rice roots and translocation within rice make understanding of GEI difficult. Greater knowledge of soil solution chemistry under field conditions may provide clues to GEI questions that remain unanswered. Breeding programs for the development of varieties with low plant As uptake need to be coupled with tolerance to As toxicity and will probably have limited success unless soil conditions are favorable for expression of genetic potentials. Growing rice in a more oxic environment, e.g., growing on raised beds instead of the traditional practice of growing on flat flooded fields, offers an opportunity to reduce both As toxicity and As concentrations in grain and straw (Duxbury and Panaullah 2007), and is consistent with a growing global need to reduce the amount of water used in rice cultivation. This approach does, however, require considerable change in the mindset of both farmers and scientists towards rice production practices.