Arsenate exposure resulted in reduced Arabidopsis root growth and branching (Figure 1). Exposure of 50 μM As (V) resulted in significantly reduced Arabidopsis root growth (Figure 1C). In addition to known relevant physiological data, the exposure study was used to determine suitable As (V) exposure for the microarray study. We noted no seed germination effects with regard to arsenate treatments, notably at 100 μM, the arsenate concentration used for transcriptomics experiments.

Forty-six genes were induced by As (V) treatment (i.e., exceeded a fold-change threshold of 1.5 and met a significance criteria of P < 0.001; FDR < 1%) as indicated by microarray analysis. The largest functional categories affected included unknown function, hydrolase, and antioxidant activity. Other functional categories affected by As (V) included genes with transferase, kinase, lyase, transporter, and binding activity [see Additional file 1; Table 1]. Alternatively, 113 genes were repressed by As (V) (i.e. exceeded a fold-change threshold of -1.5 and met a significance criteria of P < 0.001; FDR < %1), with unknown function, hydrolase, and binding activity representing the largest categories. Genes with transporter, kinase, transferase, and transcriptional regulator activity were also repressed by As (V) [see Additional file 1; Table 2]. Differentially expressed As (V)-induced and -repressed genes are listed below (Table 1 and Table 2, respectively) and complete lists of all genes affected by As (V) stress are also included as additional files [see Additional files 3 and 4]. Most interestingly, it was discovered that As (V) stress repressed transcription of many genes involved in the phosphate starvation response, and also repressed several transcriptional factors. Several genes involved in oxidative stress were also highly modulated in response to As.

SODs represented the highest ranked of both significantly induced as well as repressed genes in response to As (V) stress (Tables 1 and 2), therefore these genes presented logical primary targets for the validation of our microarray data. Results demonstrated 4.57-fold induction of a chloroplast Cu/Zn SOD (at2g28190), 2.41-fold induction of a Cu/Zn SOD (at1g08830), as well as a 3.16-fold induction of an SOD copper chaperone (at1g12520). Alternatively, Fe SOD (at4g25100) transcripts were downregulated in response to arsenic stress (-5.17-fold change). These findings were confirmed with quantitative RT-PCR (qRT-PCR) (Table 3).

Based upon our observations of transcript-level changes in SOD gene expression, we performed a nondenaturing PAGE enzyme assay of superoxide dismutase activity 16 to assess whether the observed specific changes in SOD transcript correlated with enzyme activity. This method enables the distinction between the three SOD isoenzymes found in Arabidopsis (CuZnSOD, FeSOD, and MnSOD) by using inhibitors of specific SODs. Gels were preincubated with KCN, which inhibits CuZn SOD, as well as H₂O₂, which inhibits both CuZn SOD and Fe SOD. MnSOD is resistant to both inhibitors (Figure 2). Plants were harvested from control plates containing no arsenate and treated plates containing 100 μM arsenate at seven-, ten-, and thirteen days post-germination. Irrespective of harvest date, CuZnSOD activity was strongly induced by arsenate treatment, whereas FeSOD activity was repressed, and MnSOD showed no change in activity, therefore providing sufficient evidence to confirm our microarray results. Transcription of other antioxidant genes (i.e., peroxidases, glutathione-S transferases, catalase) were indicated by our microarray experiment as affected by arsenate stress, however these genes were not included in our qRT-PCR validation.

Our microarray experiment indicated that eight different genes encoding proteins with known transcription factor activity all displayed lower expression levels in As (V)-stressed plants (Table 2). One of these transcription factors (at1g12610) encodes a member of the DREB subfamily A-1 of the ERF/AP2 transcription factor family (DDF1). One other AP2-domain-containing transcription factor (at4g34410) that encodes a member of the ERF (ethylene response factor) subfamily B-3 of the ERF/AP2 transcription factor family was also repressed in response to As (V). Two zinc finger (C2H2 type) genes (at3g46090, at3g46080) encoded a ZAT7 and a protein similar to ZAT7, respectively. Also exhibiting lower expression in As (V)-treated plants were three members of the WRKY family of transcription factors (at2g38470, at4g23810, at1g80840), WRKY33, WRKY53, and WRKY40, respectively as well as one gene encoding NAC domain containing protein 81.

A notable transcriptional trend is that As (V) stress results in repression of some genes involved in the Pi starvation response. Of particular interest, a P-type cyclin (at5g61650) that was affected by As (V)-stress shares significant homology to the PHO80 gene from yeast. We performed qRT-PCR for this gene and found that its expression was actually strongly repressed at both day 3 and day 10 (Table 3; Figure 3). Three genes that were repressed by As (V) in this study have also been reported to be repressed by Pi starvation 17. Interestingly, the three highest ranking differentially expressed genes found to be strongly induced by Pi starvation (at1g73010 > at5g20790 > at1g17710, respectively) 18, were also repressed by As (V) in our study. These genes are of particular interest on account of their unknown function. Quantitative RT-PCR confirmed that transcription of these genes was strongly repressed at both 3 day and 10 day time points. Additionally, the qRT-PCR data indicate that these genes were more repressed at day 3 than at day 10 (Table 3; Figure 3). These data also corroborated our microarray experiments, which reflect global expression ratios at 10 days post germination (Table 3).

The role that thiol groups play in arsenic detoxification has been well characterized, therefore we expected to see induction of genes involved in sulfate assimilation and metabolism in response to arsenic stress. Ferredoxin (at1g10960), a key redox protein found in the chloroplast was As (V)-induced. Expression levels for another gene involved in the sulfate reduction pathway, 5'-adenylylsulfate reductase (APR3) (at4g21190) were also elevated in response to As (V) stress. This enzyme catalyzes the reduction of APS to sulfite using glutathione as an electron donor. Although not involved in sulfate assimilation, the cysteine-rich metal-binding protein, metallothionein (MT) 1A (at1g07600) was also induced. Arabidopsis knockout mutants that were generated for class 1 MTs accumulated significantly less aboveground As, Cd, and Zn, suggesting that class 1 MTs may play a role in metal and metalloid ion translocation 19.

A wide range of genes encoding proteins involved in cell wall activities exhibit altered expression levels in response to As (V) (Table 1; Table 2). Peroxidases, which were indicated by microarray as affected by As (V) stress, are known to strengthen the cell wall in response to biotic stress via formation of lignin, extension cross-links, and dityrosine bonds 20. Additionally, As (V) affected transcription of numerous xyloglucan endotransglucosylase/hydrolases (XTHs) and glycosyl hydrolase genes (Table 1; Table 2), with the majority of these exhibiting lower expression in the presence of As (V).

Our microarray data corroborate those of Tran et al. 27, suggesting the involvement of a different NAC domain-containing transcription factor (at5g08790) in expression of FeSOD, as well as several other genes known to exhibit NAC-dependent expression. NAC proteins comprises a large gene family (> 100 members in Arabidopsis) of plant-specific transcription factors that have roles in wide-ranging processes such as development, defense, and abiotic stress response 28. Microarray experiments were carried out on NAC-overexpression Arabidopsis mutants to discover genes exhibiting dependence on NAC transcription factors for transcription 27. We speculate that repression of NAC81 (at5g08790) in As (V)-stressed Arabidopsis may be responsible for the observed repression of FeSOD (at4g25100), ferritin 1 (FER 1) (at5g01600), XTH15 (at4g14130), XTH24 (at4g30270), erd1 ATP-dependent Clp protease ATP-binding subunit (at5g51070), and a branched-chain amino acid amino transferase 2 (at1g10070), as these genes were reported as exhibiting NAC-dependent expression 27.

Although phosphate is undoubtedly one of the most biologically important nutrients, its availability in soils is quite low. Therefore, plants have evolved mechanisms to maximize Pi accessibility/availability, such as increased root hair growth, lateral root branching, and induction of phosphate transporters and phosphatases 29. Certain phosphate starvation-induced genes have evolved to release phosphate from plasma membranes by hydrolyzing phospholipids under conditions of low Pi availability, as phospholipids comprise a major Pi pool in planta 30. Conversion from phospholipids to galactolipids is one such strategy and can result from the activity of monogalactosyldiacylglycerol (MGDG) synthase or digalactosyl diacylglycerol (DGDG) synthase 31. Arabidopsis plants expressing MGD2 and MGD3 promoter-GUS fusion constructs showed that under Pi starvation, MGD3::GUS was expressed in apices of serrated edges (hydathodes) and in the lateral root branch 31. Through investigation of Arabidopsis MGDG synthase gene expression under Pi starvation, these authors showed that global changes in plant membranes under Pi deprivation are tightly regulated by Pi signaling and that signal transduction through a Pi-sensing mechanism is responsible for regulating MGDG synthase gene expression 31. We report here that the expression of MGD3 (at2g11810) is lower in As (V)-treated Arabidopsis at 3 days and 10 days (Table 3; Figure 3). Therefore, it is conceivable that our observations may either reflect a Pi/As (V) sensing mechanism or simply the lower number of lateral roots in As (V)-stressed plants (Figure 1). SENESCENCE RELATED GENE 3 (SRG3; at3g02040), a glycerophosphoryl diester phosphor-diesterase, is believed to participate in processes similar to those of the MGDG synthase genes 32SRG3 had lower transcript abundance in As (V)-treated plants in our microarray study (along with other senescence-associated proteins) (Table 2), as well as, in 3 day and 10 day As (V)-treated plants (Table 3; Figure 3). A type 5 acid phosphatase (ACP5; at3g17790) was also repressed in our As (V)-treated plants as indicated by microarray (Table 2) and was strongly repressed in our qRT-PCR validation experiments at both 3 day and 10 day time points (Table 3; Figure 3). In Arabidopsis, ACP5 has been shown to be induced by H₂O₂, but not by paraquat or salicylic acid and is thought to be involved in both phosphate mobilization and in the metabolism of reactive oxygen species 33. In contrast, ACP5 was strongly repressed by As (V) despite elevated SOD levels, which generate H₂O₂. Therefore, further study is required to determine the specific cause of As (V)-mediated ACP5 repression.

Recent investigations into the genome-scale transcriptional changes to phosphate deprivation in Arabidopsis have elucidated a broad range of genes involved in phosphate metabolism 1718. Our microarray data suggested that many genes repressed by As (V) stress have been reported by others 1718 to be induced in response to Pi deprivation in Arabidopsis thaliana. Because As (V) behaves as a phosphate analog, it is likely that this observation can be explained by a saturation effect of the phosphate analog, As (V), thereby misleading metabolic and regulatory perception of the toxic metalloid as an abundant supply of Pi. However, arsenate likely disrupts critical biological processes that involve reversible phosphorylation, as well as pathways for phosphate signaling, but even under arsenate stress, Arabidopsis accumulates much higher concentrations of As in the root than is translocated to the shoot. In another study, when wild-type (Columbia ecotype) Arabidopsis plants were grown on 100 μM sodium arsenate for 3 weeks, low concentrations of arsenic were accumulated in the shoot, whereas high concentrations of arsenic were observed in roots 34. However, when the arsenate reductase homolog (ACR2) was silenced, arsenate was translocated to the shoot at concentrations that classified as hyperaccumulation 34. Nevertheless, the signaling mechanisms by which plants distinguish between As (V) and phosphate are unknown and other mechanisms of As detoxification and storage besides the well documented phytochelatin response 9101112 may exist.

In order to confirm the observation that As (V) stress represses genes involved in phosphate starvation/acquisition, we performed qRT-PCR on some of the more interesting candidates (Table 3; Figure 3). We are particularly interested in elucidating pathways involved in As (V) signaling in plants. The P-type cyclin (at5g61650) that was affected by As (V) (Table 3; Figure 3) shares significant homology to the PHO80 gene from yeast. Cyclins bind and activate cyclin-dependent kinases, which play key roles in cell division via phosphorylation of critical substrates, such as the retinoblastoma protein, transcription factors, nuclear laminar proteins, and histones 35. Interestingly, it was demonstrated that expression of this cyclin from Arabidopsis restored the phosphate signaling pathway in a PHO80-deficient yeast mutant, suggesting a putative key Pi signaling role 36.

Protein kinases play crucial roles in signal transduction pathways in all eukaryotes 37. At3g08720 (ATPK19) is one of two nearly identical kinase genes in Arabidopsis that encode for proteins that share high sequence homology with the mammalian 40S ribosomal protein kinases S6K1 and S6K2 38. ATPK19 was demonstrated to be the functional plant homolog of mammalian p70s6k when ectopic expression of this gene specifically phosphorylated ribosomal protein S6 derived from either plant or animal 39. ATPK19 has recently been implicated as a crucial nodal point in a network evolved for integrating stress signals with plant growth regulation 40. Lower expression levels observed for ATPK19 in As (V)-treated plants, which was most severe at day 3 (Table 3; Figure 3), lends us to conclude that As (V) stress may suppress plant growth through the downregulation of this growth-regulating kinase, possibly as a result of the chemical similarity between As (V) and phosphate. Alternatively, the downregulation of ATPK19 may result from the more general stress responses imposed by the toxic metalloid (e.g. oxidative stress, sulfhydryl group binding, etc.).

Our results are in agreement with the recently proposed ideas of Catarecha et al. 41 who studied an Arabidopsis mutant that displayed enhanced arsenic accumulation. These authors identified a Pi transporter (PHT1;1) mutant with a decreased rate of As (V) uptake and increased As (V) accumulation. By comparing gene expression of the mutant with wild-type plants, it was shown that in Arabidopsis, As (V) rapidly repressed genes involved in the Pi starvation response and induced the expression of other As (V)-responsive genes 41. Interestingly, the repression of Pi starvation genes was shown to be specific for As (V), whereas the As (V)-induced genes were also induced by As (III). A model resulted that suggests arsenic acts via two separate signaling pathways 41. Because of the chemical similarity of As (V) and Pi, As (V) fools the Pi sensor, thus initiating the repression of the Pi starvation response. Although our microarray experiments did not detect differential expression of any high-affinity Pi transporter, which may be due to differences in experimental approach, Catarecha et al. 41 illustrated the high sensitivity of the Pi transporter, PHT1;1, to As (V) and suggested that plants have evolved an As (V) sensing system whereby As (V) and Pi signaling pathways oppose each other to protect the plant from arsenic toxicity. Based on our results, it is conceivable that the P-type cyclin (at5g61650) and ATPK19 (at3g08720) may be involved in As (V) sensing, but further study is required to confirm this finding.

Our comparison of As (V)-repressed genes that have also been shown to be induced by Pi deprivation elucidate some promising candidates for future studies. For example, we are particularly interested in genes with unknown function that are strongly induced in both roots and leaves by Pi starvation (i.e. at1g73010; at1g17710; at2g04460; at5g20790; 17, supplemental data; 18). Both at1g73010 and at1g17710 are described as phosphoric monoester hydrolases (see Availability and requirements section for URL), but to our knowledge, these have not been studied in this regard. Most recently, a study described gene networks for the Arabidopsis transcriptome based on the graphical Gaussian model of global-scale transcriptional studies 32. In a constructed subnetwork of genes involved in phosphate starvation, SRG3, at1g73010, at3g17790 (ACP5), at2g11810, and at5g20790 were all closely linked, suggesting their critical roles in phosphate metabolism in Arabidopsis 32. Based on their strong induction in response to Pi starvation 1718, it is reasonable to conceive that at1g73010 and at1g17710 have evolved as Pi scavengers for increasing Pi availability. We have confirmed the downregulation of these genes in response to As (V) at both 3 and 10 day time points (Table 3; Figure 3). In this study, at2g04460 transcript levels were strongly repressed at 3 and 10 day time points, whereas at5g20790 was repressed at day 3 and day 10 (Table 3; Figure 3). Interestingly, at2g04460 encodes for a putative retroelement pol polyprotein that has been reported as highly expressed in salt overly sensitive (sos) Arabidopsis mutants 42. Because the function of these two Pi starvation-induced genes is unknown, these putative gene candidates may provide opportunities for gaining insight into As (V)/Pi dynamics in Arabidopsis thaliana.

The data presented here have led to the development of new hypotheses for future research. The potential antagonistic effects of various arsenate and Pi concentrations on the expression of the aforementioned genes in Arabidopsis are poorly understood. Additionally, the efficiency of arsenate reduction and subsequent detoxification via phytochelatins or glutathione is poorly understood. Under conditions of arsenate stress (i.e., 100 μM), perhaps the cellular concentrations of arsenate surpass those that may be efficiently reduced by glutathione or arsenate reductase, thus allowing free arsenate to interfere with biological reactions that involve phosphate. Additionally, the selected arsenate concentration to employ in this study could have resulted in free arsenite after reduction in vivo that would likely have deleterious consequences. Therefore, it is reasonable to conceive that the observed transcriptional responses, as well as the impaired phenotype seen in this study, may be reflective of either arsenate, arsenite, or both.

Seeds of Arabidopsis thaliana ecotype Columbia plants were surface sterilized and plated on agar-solidified MS culture medium supplemented with B5 vitamins, 10% sucrose, 2% Gelrite^®, pH 5.8. Phosphate is supplied as 1.25 mM KH₂PO₄ in the culture medium. Arsenic-treated plates were supplemented with 100 μM potassium arsenate (Sigma) according to a previously determined sub-lethal growth response curve. Plates were cold stratified at 4°C for 24 hrs and then placed in a growth chamber at 25°C under a 16 hr photoperiod. At each time point (3 d, 10 d), 2 g of whole plant material (shoots + roots) was harvested from each plate, frozen in liquid nitrogen, and subjected to RNA isolation using Trizol^® reagent (Invitrogen, Carlsbad, CA) according to manufacturer's protocol. A total of three biological replicates were assayed (3 control, 3 treated) where each pooled 2 g sample represented a single biological replicate.

Total RNA from six biological replicates were purified using RNeasy MiniElute columns (Qiagen, Valencia, CA). A total of 1.25 μg of purified total RNA was subjected to Aminoallyl Message Amp II kit (Ambion, Austin, TX) first strand cDNA synthesis, second strand synthesis, and in vitro transcription for amplified RNA (aRNA) synthesis. aRNA was purified according to manufacturers protocol (Ambion, Austin, TX) and quantified using a Nanodrop spectrophotometer. Two 4 μg samples of aRNA were labeled with Cy3 and Cy5 monoreactive dyes (Amersham Pharmacia, Pittsburgh, PA) in order to conduct a dye swap technical replicate for each biological replicate. Each aRNA sample was brought to dryness in a Speedvac and dissolved in 5 μL of 0.2 M NaHCO₃ buffer. Five microliters of Cy3 or Cy5 (in DMSO) was added to each sample and incubated for 2 hrs in the dark at RT. Labeled aRNA was purified according to kit instructions (Ambion, Austin, TX) and quantified using the Nanodrop spectrophotometer. One-hundred pmol Cy3- and Cy5-labeled aRNA targets were denatured by incubating at 65°C for 5 min and added to a hybridization mix containing 9 μl 20× SSC, 5.4 μl Liquid Block (Amersham Pharmacia, Pittsburgh, PA), and 3.6 μl 2% SDS for a 90 μl total volume.

Microarrays comprised of 70-mer oligonucleotides obtained from the University of Arizona (see Availability and requirements section for URL) were immobilized by rehydrating the slide over a 50°C waterbath for 10 s and snap drying on a 65°C heating block for 5 s for a total of four times. Slides were UV-crosslinked at 180 mJ in a UV cross-linker (Stratagene, La Jolla, CA). The slides were then washed in 1% SDS, dipped in 100% EtOH five times followed by 3 min shaking. Slides were spun dry at 1000 rpm for 2 minutes and immediately placed in a light-proof box. The 90 μl hybridization mix was pipetted onto a microarray slide underneath a lifterslip (Lifterslip, Portsmouth, NH) and placed in a hybridization chamber (Corning, Corning, NY) overnight at 55°C. After hybridization, slides were washed in 2× SSC, 0.5% SDS for 5 minutes at 55°C, 0.5× SSC for 5 minutes at room temperature, and 0.05× SSC for 5 minutes at room temperature. Slides were then spun dry at 1000 rpm in a Sorvall centrifuge and scanned with a GenePix 4000B scanner (Axon Instruments, Inc., Union City, CA). The intensity variation was removed by fitting a loess regression using SAS 9.1 (SAS, Cary, NC). Data were log-2 transformed and statistically analyzed using rank product statistics as described by 43 to identify differentially expressed genes. Bioconductor Rank Prod package was used to perform the rank product analysis 4445. Significantly different genes reported in this study exhibited P < 0.001, as designated by the rank product analysis. The false discovery rate (FDR) 46 value obtained was based on 10,000 random permutations. Since 10,000 random permutations was very computer intensive, 1000 random permutations were performed 10 different times each time starting with a different random seed number and the average FDR value calculated was used for further analysis. The genes that had FDR values less than or equal to 0.01 were considered as differentially expressed. Data for all microarray experiments were submitted to the NCBI GEO microarray database and can be viewed under the accession GSE10425.

Global gene expression profiling comparing arsenate-treated Arabidopsis plants with control was carried out to better understand the mechanisms of plant response to arsenate stress and to identify genes involved in arsenic metabolism. For microarray data quality control, we examined both dye dependent effects and distribution of the ratio after normalization. We have included an additional file that illustrates the quality of microarray experiments, as well as the overall gene expression pattern [see Additional file 2]. Additional file 2a shows the normalized M vs. A plot, which was generated as a scatter plot of log intensity ratios M = log₂ (R/G) versus average log intensities A = log₂(R*G)/2, where R and G represent the fluorescence intensities in the Cy3 and Cy5 channels, respectively 47. As shown by the figure, Loess normalization effectively removed dye dependent effects in the microarray and rendered evenly distributed ratios across all signal intensities. The histogram suggests a normal distribution of the logarithm 2-based transformed ratio [see Additional file 2b]. Overall, the microarray experiments generated high quality data without significant dye-dependent effects and skewness of ratio distribution.

Gene ontology annotations were translated from microarray data using the GO annotations bioinformatics tool available at The Arabidopsis Information Resource Web site http://www.arabidopsis.org/tools/ where results were based on molecular function.

Total RNA was extracted from Arabidopsis thaliana ecotype Columbia grown for ten days as described for the microarray experiment. Five micrograms of total RNA was reverse-transcribed with oligo(dT)₂₀ primers using the Superscript III first-strand cDNA synthesis kit (Invitrogen, Carlsbad, CA). RT PCR was performed using the ABI 7000 Sequence Detection System (Applied Biosystems, Foster City, CA). PCR was performed in a 15 μl reaction volume containing Power Sybr^® PCR mix (Applied Biosystems, Foster City, CA) and gene-specific primers were designed with PrimerExpress software. Actin was used as the reference gene, and the primer sequences for Arabidopsis actin gene were AGTGGTCGTACAACCGGTATTGT (F) and GAGGAAGAGCATTCCCCTCGTA (R). After the RT PCR experiment, Ct number was extracted for both reference gene and target gene with auto baseline and manual threshold.

The cluster analysis was conducted with MultiExperiment viewer Version 4.0 (TIGR, Rockville, MD) with logarithm 2 transformed ratio of treated vs. control samples from real-time PCR. The complete linkage hierarchical cluster was used to cluster the genes only. The color scheme is as shown in the figure, with repressed genes shown as green and red color indicating induced genes.

Total soluble protein was extracted from whole Arabidopsis plants (root + shoot) grown on plates as described above that were harvested at each respective time point. Total soluble protein was quantified by the method of Bradford 48 using BSA as a standard and 50 μg samples were loaded. Bovine SOD (Sigma) was used in each gel to serve as a positive control for SOD activity. Following electrophoretic separation on a 10% non-denaturing polyacrylamide gel, SOD activity was determined as described by Beauchamp and Fridovich (1971) and modified by Azevedo et al. 49. The gels were rinsed with DDI water and incubated in the dark for 30 min at room temperature in a reaction mixture containing 50 mM potassium phosphate buffer (pH 7.8), 1 mM EDTA, 0.05 mM riboflavin, 0.1 mM nitroblue tetrazolium and 0.3% (v/v) TEMED. Following incubation, gels were rinsed with DDI water and illuminated in water until SOD bands were visible. The gels were then immersed in a 6% (v/v) acetic acid solution to stop the reaction. To confirm specificity of Cu/Zn-SOD activity, H₂0₂ and KCN were used as inhibitors as described by Azevedo et al. 49 and modified by Vitoria et al. 50. Mn-SOD is resistant to both inhibitors, Fe-SOD is resistant to KCN and inhibited by H₂0₂, and Cu/Zn-SOD is inhibited by both inhibitors, thus allowing classification of SOD activity. Prior to SOD staining, gels containing lanes in triplicate were cut into three parts; one gel was treated as described above, the second and third parts were incubated for 20 min in 100 mM potassium phosphate buffer (pH 7.8) containing either 2 mM KCN or 5 mM H₂O₂, respectively. Following incubation, gels were rinsed with DDI water and then stained for SOD activity.