SAGE Analysis of Transcriptome Responses in Arabidopsis Roots Exposed to 2,4,6-Trinitrotoluene

Abstract

Serial analysis of gene expression was used to profile transcript levels in Arabidopsis roots and assess their responses to 2,4,6-trinitrotoluene (TNT) exposure. SAGE libraries representing control and TNT-exposed seedling root transcripts were constructed, and each was sequenced to a depth of roughly 32,000 tags. More than 19,000 unique tags were identified overall. The second most highly induced tag (27-fold increase) represented a glutathione S-transferase. Cytochrome P450 enzymes, as well as an ABC transporter and a probable nitroreductase, were highly induced by TNT exposure. Analyses also revealed an oxidative stress response upon TNT exposure. Although some increases were anticipated in light of current models for xenobiotic metabolism in plants, evidence for unsuspected conjugation pathways was also noted. Identifying transcriptome-level responses to TNT exposure will better define the metabolic pathways plants use to detoxify this xenobiotic compound, which should help improve phytoremediation strategies directed at TNT and other nitroaromatic compounds.

Soil and groundwater at sites throughout the United States and Europe were contaminated in the past century by manufacturing, processing, and storage of explosives, such as 2,4,6-trinitrotoluene (TNT; Walsh et al., 1993). Unlike many other nitroaromatic compounds, including pesticides and various feed-stock chemicals, the energetic nitroaromatics and heterocyclic nitroamines (Fig. 1) are highly resistant to degradation and may persist in the environment for decades. Certain plant species have the ability to accumulate TNT from their surroundings and thus offer a potential means for removing these compounds from the environment (Hannink et al., 2002). Unfortunately, few of these species are capable of tolerating the contamination levels common to sites most in need of remediation. Transgenic plants have been developed with enhanced abilities to tolerate and remove TNT from soil under laboratory conditions (French et al., 1999; Hannink et al., 2001). However, lack of information on the biochemical mechanisms involved in TNT uptake and metabolism limits our ability to modify and adapt plants specifically for this task.

Figure 1.

Chemical structures of nitroaromatic and heterocyclic nitroamine explosives. The nitroaromatic explosives of greatest concern for environmental contamination, TNT, RDX (Royal Demolition Explosive), and HMX (High Melting Explosive), are depicted.

To obtain a more complete picture of the metabolic processes plants employ to cope with nitroaromatic agents, serial analysis of gene expression (SAGE; Velculescu et al., 1995, 1997) was used to identify transcriptome-level responses in Arabidopsis seedling roots exposed to TNT. Although this technique has been widely used to study gene expression changes in human cancers (Riggins and Strausberg, 2001), only two published reports describe the use of SAGE in plants: a study of gene expression in rice (Oryza sativa) seedlings (Matsumura et al., 1999) and an examination of wood formation in loblolly pine (Lorenz and Dean, 2002). The identification of plant genes and pathways responding to TNT exposure at the transcriptional level should facilitate a more reasoned approach to the development of plants better suited for use in phytoremediation of TNT-contaminated sites.

RESULTS

SAGE Library Characterization and Comparison of Tags to the Arabidopsis Genome

SAGE libraries representing transcripts expressed in Arabidopsis root tissues grown in the presence (15 mg L^–1) or absence of TNT were sequenced to characterize about 30,000 tags from each. The data sets from these analyses are available through the National Center for Biotechnology Information Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/). Of the 32,203 tags characterized from the library for TNT-treated tissue, 12,005 represented unique transcripts, but 7,900 of these were singletons. Similarly, 12,721 unique tags were encountered among the 32,104 characterized in the control library, with 8,322 of these representing singletons. A double-reciprocal plot of the total unique tags identified versus the total number of tags sequenced after each sequencing run was used to estimate the rate of transcript discovery (Fig. 2). From this, transcriptome sizes of approximately 21,000 and 15,000 were estimated for control and TNT-treated root tissues, respectively. Only 25% of all unique tags (5,084 of 19,640) were detected in both libraries.

Figure 2.

Estimates of transcriptome size in Arabidopsis roots. Double-reciprocal plots of new tag discovery versus total tags sequenced are shown for the TNT-treated (top) and control (bottom) root libraries. The estimated transcriptome size was taken as the 1/y intercept from the linear regression depicted in each graph.

Of the 1,045 most abundant tags in both SAGE libraries (which included all tags seen 10 or more times among the 64,176 tags characterized) about 70% (739), could be matched to a single model gene in the Arabidopsis Genome Initiative (AGI) database. Most of the remaining tags (233) matched sequences found one or more times among all Arabidopsis sequences deposited in GenBank. Many such tags were found in expressed sequence tags (ESTs) but were not positioned adjacent to the CATG sequence closest to the 3′ end of the transcript. This suggests that they might have been generated from alternatively spliced transcripts. A good example of this was the fourth most abundant tag in the control tissue library (AGGTCTTGGT, counted 134 times). This sequence appears only one time in the Arabidopsis genome and falls immediately 3′ of the penultimate CATG site on the annotated transcript of gene At3g09260 (β-glucosidase). The full-length transcript for this gene annotated in the AGI database corresponded to the second most abundant tag in the control library (ATTTGCCAGA, counted 286 times) and the fourth most abundant tag in the TNT treatment library (counted 259 times).

Of the remaining unidentified tags in the top 1,045, one that did not correspond to anything listed in the AGI database (AGTAACGATA) matched a sequence found on the Arabidopsis mitochondrial genome, and numerous ESTs incorporating this sequence have been deposited in GenBank. About 5% of the tags (57) matched more than one model gene, whereas a few tags (14) contained at their 3′ end a contiguous stretch of “A” residues, suggesting that they incorporated part of a poly(A) tail. Two tags, one of which (TCCCCGTACA) was the 37th most abundant tag overall, could not be matched to any Arabidopsis sequences in GenBank.

Identification of Differentially Expressed Transcripts

TNT treatment induced an apparent increase of at least 5-fold in 242 tags, whereas 287 tags decreased in abundance at least 5-fold in response to this treatment. For tags that were relatively abundant in one library but not observed in the other, a minimum change in expression level was estimated by assuming that a single copy of the tag was found in the library from which it was absent. Several of the tags displaying the greatest induction in response to TNT exposure represented gene products known to be involved in plant responses to oxidative stress, such as monodehydroascorbate reductase and glutathione (GSH)-dependent dehydroascorbate reductase (Table I). The identity of the tag most highly induced by TNT remains uncertain. Its sequence appears twice on chromosome 5, and in one of these instances, the tag lies in the middle of a computationally predicted transcript that has never been isolated. However, a third occurrence of the sequence resides on chromosome 2 at the penultimate CATG site of the transcript for LKP2 (At2g18915), a signaling protein involved in the Arabidopsis circadian clock (Schultz et al., 2001). The evidence suggests the most likely source of this tag to be a splice variant of the LKP2 transcript. In contrast, the second most highly induced tag, which increased nearly 28-fold in TNT-treated roots, definitely represented a glutathione S-transferase (GST). Several of the induced tags represented cytochrome P450s, a large family of enzymes used by plants and other organisms in the detoxification of xenobiotic compounds (Table II). As further indication of the stress imposed upon the seedlings by TNT exposure, tags for various transporter proteins, transcription factors, signal cascade proteins, and heat shock proteins were among those most highly induced by TNT exposure. Other strongly induced tags representing gene products known to be involved in detoxification reactions or in the protection of plant cells against oxidative stress are noted in Table III. Prominent among the enzymes likely to protect against oxidative stress are quinone reductase, ascorbate peroxidase, peptide Met sulfoxide reductase, phospholipid-hydroperoxide GSH peroxidase, and GSH reductase.

Table I.

SAGE tags induced at least 10-fold by exposure to TNT

Tag Sequence	Tag Abundance		-Fold Increasea	P-Chanceb	Locus	Annotation
	TNT	Control	TNT/ Control
GTGAGTTTGA	30	0	>30.0	0	At2g18915c	F-box protein LKP2/ADO2
CCAAATTCTG	55	2	27.5	0	At1g17170	GST, putative
TAGCCAATTA	72	3	24.0	0	At3g44300	Nitrilase 2
GGAGTTTGTA	23	1	23.0	0	At3g27740c	Carbamoyl phosphate synthetase (small subunit)
ATGTTTCGCG	21	1	21.0	6.68E-06	At5g13750	Transporter-like protein
GGAGAAGTCC	20	1	20.0	1.67E-05	At1g60940	Serine/threonine-protein kinase, putative
GCAATTCTAC	35	2	17.5	0	At5g03630	Monodehydroascorbate reductase, putative
AGAAGTTTAT	17	1	17.0	8.67E-05	At5g08790c	No apical meristem-like protein
TGAGTTTCAA	17	1	17.0	8.67E-05	At4g01870	Expressed protein
GGTTAGTCGA	15	1	15.0	3.13E-04	At3g41950c	Unprocessed transcript? (AY090986)
AGAGAAAGTG	14	0	>14.0	5.34E-05	At3g28740	Cytochrome P450, putative (CYP81D11-A-TYPE)
CGGGGAAAAA	13	1	13.0	9.70E-04	?	Polyadenylation sequence?
AGTTGAGTTC	38	3	12.7	0	At4g19880	Expressed protein
GTGAAGTTTG	12	0	>12.0	1.90E-04	At3g45270	Putative protein (similar to AC-like transposases)
GGAAAAGGTG	23	2	11.5	2.34E-05	At3g21720	Isocitrate lyase, putative
ACCAAAATTG	11	0	>11.0	5.10E-04	At2g29490	GST, putative
CCACAGTTTT	30	3	10.0	0	At1g78080	AP2 domain protein, putative (RAP2-like transcription factor, TINY)
AACGCAGAAA	10	0	>10.0	1.05E-03	At1g01550	Expressed protein
CAGGATGTGT	10	0	>10.0	1.05E-03	At1g05680	Indole-3-acetate beta-glucosyltransferase, putative
GCACTCTTGA	10	0	>10.0	1.05E-03	At2g28570	Expressed protein (protein-Tyr kinase motif)
CTTCTCTAGT	10	1	10.0	6.37E-03	At1g75270	GSH-dependent dehydroascorbate reductase, putative
CTTGTCCTCA	10	1	10.0	6.37E-03	At1g76680	12-Oxophytodienoate reductase (OPR1)

^a For the calculation of expression ratios, a value of 1 was substituted where tag counts were zero.

^b P-chance values are averages of three Monte Carlo simulations.

^c Likely alternative splicing variant.

Table II.

Cytochrome P450 genes induced by exposure to TNT

Tag Sequence	Tag Abundance		-Fold Increasea	P-Chanceb	Locus	Classificationc	Typec
	TNT	Control	TNT/Control
AGAGAAAGTG	14	0	>14.0	5.34E-05	At3g28740	CYP81D11 (putative)	A
GAAAAGATTT	5	0	>5.0	3.10E-02	At2g30750	CYP71A12	A
GCTGAGAGAC	5	0	>5.0	3.10E-02	At4g22690	CYP706A2 (like protein)	A
					At4g22710
ATGCGAAGCT	41	9	4.6	3.34E-06	At2g30490	CYP73A5 (type enzyme) trans-cinnamate 4-mono-oxygenase / cinnamate-4-hydroxylase (C4H)	A
AAGCATCCGC	4	0	>4.0	6.08E-02	At1g64900	CYP89A6 (hypothetical)	A
					At1g64940
					At1g64950

^a For the calculation of expression ratios, a value of 1 was substituted where tag counts were zero.

^b P-chance values are averages of three Monte Carlo simulations.

^c Cytochrome P450 classification as described by Paquette et al. (2000).

Table III.

TNT-induced SAGE tags representing potential detoxification pathway components

Tag Sequence	Tag Abundance		-Fold Increasea	P-Chanceb	Locus	Annotation
	TNT	Control	TNT/Control
TTGAGAAATT	9	1	9.0	1.06E-02	At5g54500c	Quinone reductase, putative
ATTCTGAGAA	9	1	9.0	2.04E-03	At1g78340	GST, putative
TGATGAGTTT	94	11	8.5	0	At3g09390	Metallothionein-related protein
TGGCGGATTA	40	5	8.0	0	At4g19880c	Expressed protein
AAGATCCAAG	8	1	8.0	1.96E-02	At2g18160c	G-box binding bZIP transcription factor
TGCAAGTTAT	8	1	8.0	1.96E-02	At3g28480	Prolyl 4-hydroxylase, putative
TAGAATTCTC	7	1	7.0	3.50E-02	At4g03430	Putative pre-mRNA splicing factor
TGATTCAAAA	7	1	7.0	3.50E-02	At2g16680c	Reverse transcriptase (RNA-dependent DNA polymerase), putative
AAACTGTTTG	7	1	7.0	8.50E-03	At5g48930c	Anthranilate N-benzoyltransferase
TCACTCCTAT	6	1	6.0	6.52E-02	At3g44300c	Nitrilase 2
CAAATCAGTT	33	6	5.5	1.34E-05	At1g76930	Extensin
TCTCGAACCT	11	2	5.5	1.17E-02	At5g20830	Suc-UDP glucosyltransferase
GCTGTTTTTG	155	30	5.2	0	At1g07890	l-Ascorbate peroxidase
TGTTTGGCTG	5	0	>5.0	3.10E-02	At3g58500	Phosphoprotein phosphatase 2A isoform
CCAATTAGTC	10	2	5.0	2.08E-02	At3g53480	ABC transporter-like protein
GAAACGCTCA	10	2	5.0	2.08E-02	At5g07460	Peptide methionine sulfoxide reductase-like protein
GTTTCGAGAT	47	10	4.7	0	At4g11600c	Phospholipid hydroperoxide glutathione peroxidase
AAAACTCGGT	9	2	4.5	3.39E-02	At3g24170	Glutathione reductase, cytosolic
CTTGGTGCAA	11	3	3.7	3.07E-02	At2g44350	Citrate synthase
ACGAAGGTCG	9	3	3.0	7.08E-02	At3g44320c	Nitrilase 3
TAACTTGTGC	20	7	2.9	1.06E-02	At1g51680	4-Coumarate:CoA ligase
GAAACTTAAA	31	11	2.8	1.45E-03	At4g30170	Peroxidase ATP8a

^a For the calculation of expression ratios, a value of 1 was substituted where tag counts were zero.

^b P Chance values are averages of three Monte Carlo simulations.

^c Likely alternative splicing variant.

Tags that showed the greatest decrease upon exposure to TNT are listed in Table IV. The largest effect was seen for a tag representing a lipid transfer protein (LTP) family member (At4g12550) that has been associated with lateral root formation in response to auxin (AIR1A; Neuteboom et al., 1999). An overlapping transcript of this gene (At4g12445, AIR1B) represented by a different SAGE tag shows the fourth most precipitous decline in response to TNT, whereas a related member of the LTP family (At4g12520, pEARLI-1 homolog) was also severely repressed in response to the TNT treatment (Table IV). TNT exposure also depressed levels of tags representing three major latex protein homologs, another group of widely distributed proteins whose functions are as yet unknown (Stromvik et al., 1999). Other depressed tags included those for various transcription factors, membrane channels, cytoskeletal elements, and ribosomal proteins.

Table IV.

SAGE tags repressed at least 14-fold by exposure to TNT

Tag Sequence	Tag Abundance		-Fold Decreasea	P-Chanceb	Locus	Annotation
	Control	TNT	Control/ TNT
TTTTCTTACC	85	0	>85.0	0	At4g12550	Protease inhibitor/seed storage/LTP family (AIR1A)
TATCCTTGTT	52	1	52.0	0	At2g01520	Major latex protein-related
TTGTATGTTT	45	0	>45.0	0	At1g70850	Bet vl allergen family (major latex protein-related)
TTTTCTTATC	37	0	>37.0	0	At4g12545	Protease inhibitor/seed storage/LTP family (AIR1B)
GACCAACCAC	29	1	29.0	0	At2g36830	Tonoplast intrinsic protein (γ)
TGTCTTAGCT	29	1	29.0	0	At1g09690	Putative 60S ribosomal protein L21
TTTATGCTTT	50	2	25.0	0	At4g22212	Expressed protein
TGTTTATTTT	24	1	24.0	0	At1g01620	Plasma membrane intrinsic protein 1c
					At5g23710	Unknown protein
TTTTCTTCTT	23	1	23.0	3.34E-06	At3g54580	Pro-rich protein family
TTACAATAAC	21	1	21.0	0	At5g19510	Elongation factor 1B α-subunit
GGAACATATA	20	0	>20.0	0	At4g38740	Peptidylprolyl isomerase ROC1
GTGTTGTATG	20	0	>20.0	0	At1g14960	Major latex protein-related
TTGGTTATGT	20	0	>20.0	0	At5g56540	Arabinogalactan-protein (AGP14)
AACCCGGCCA	16	0	>16.0	2.34E-05	At4g17340	Major intrinsic protein family
TTTTTCTTTG	15	0	>15.0	2.00E-05	At4g12520c	Protease inhibitor/seed storage/LTP family (pEARLI-1 homolog)
TCTTTGTCTT	15	1	15.0	1.87E-04	At1g21100	O-methyltransferase 1
AGTTTATCAC	14	0	>14.0	4.00E-05	At5g04750c	F1F0-ATPase inhibitor-like protein
TTATCTCTCT	14	0	>14.0	4.00E-05	At1g04820	Tubulin α-2/α-4 chain
TTTTGCTATC	56	4	14.0	0	At5g26260	Expressed protein
CAACATTGTA	42	3	14.0	0	At5g14200	3-Isopropylmalate dehydrogenase
TTTCTGGTAA	14	1	14.0	4.90E-04	At4g14960	Tubulin α-6 chain (TUA6)

^a For the calculation of expression ratios, a value of 1 was substituted where tag counts were zero.

^b P-chance values are averages of three Monte Carlo simulations.

^c Likely alternative splicing variant.

Validation of SAGE Results Using Quantitative PCR

SAGE has previously been shown to provide an accurate reflection of gene expression levels for medium- and high-abundance transcripts (Evans et al., 2002), but the induction or repression of selected Arabidopsis transcripts in response to TNT exposure was independently verified in this study using real-time quantitative PCR. Genes that SAGE analysis suggested were induced (At3g28740; cytochrome P450), repressed (At2g36830; tonoplast intrinsic protein), or remained unaffected (At2g39460; 60S ribosomal protein L23A) by TNT exposure were tested. As shown in Table V, the quantitative PCR data showed general agreement with the SAGE data. With respect to apparent discrepancies between the values determined by the two techniques, it should be noted that the value determined by quantitative PCR is a logarithmic, rather than a linear function of the starting number of templates present. Thus, ratios of expression differences are not generated by simple division of threshold cycle (C_t) values, and even slight differences in the PCR reaction conditions can have strong impact on the C_t values. With respect to SAGE, due to sampling effects, differential expression can be strongly affected by the extent to which sequencing is carried out. Thus, the SAGE-determined value for P450 (At3g28740) induction is significantly underestimated because of the constraints placed on the depth of sequencing in the SAGE libraries. These factors are likely the most significant sources for the slight discrepancies noted in the data produced by the two techniques.

Table V.

Comparison of transcript expression levels as determined by SAGE and qPCR

Amplicon (Model Gene)	Control		TNT		Ratio TNT/Control
	qPCRa	SAGE	qPCR	SAGE	qPCR	SAGE
Cytochrome P450 (At3g28740)	8.6 ± 1.8	0	397 ± 2.8	14	46	>14
Ribosomal Protein L23A (At2g39460)	122 ± 1.2	31	232 ± 20	34	1.9	1.1
Tonoplast intrinsic protein (At2g36830)	116 ± 17	29	10.5 ± 1.4	1	0.09	0.03

^a Average values for duplicate qPCR measurements ± the sd.

DISCUSSION

A major goal of this study was to identify plant-specific enzymes and metabolic pathways that might previously have been overlooked for their importance in conferring tolerance to TNT. To avoid confusion from microbial influences and to minimize variation resulting from environmental fluctuations, this study used root tissues from seedlings grown in sterile liquid culture. The tolerance of Arabidopsis to TNT under these conditions was comparable with that seen in other plants species grown under sterile culture conditions (Hannink et al., 2001). For example, TNT was shown to induce chlorosis and cell death in cultures of Anabena sp. at a concentration of 10 mg L^–1 (Pavlostathis and Jackson, 1999), whereas cell suspension cultures of Datura innoxia were capable of tolerating TNT at up to 29 mg L^–1 without an effect on growth (Lucero et al., 1999). Although the liquid culture system used in this study exposed all portions of the Arabidopsis seedlings to TNT, only the root tissues were sampled for SAGE analyses because under conditions of normal terrestrial growth TNT and its derivatives tend to remain associated with root tissues (Cataldo et al., 1989). Thus, efforts to improve plant tolerance to TNT would seem most likely to benefit from detailed study of the metabolic responses in these tissues.

SAGE Analysis of the Arabidopsis Transcriptome

The Arabidopsis genome has been estimated to contain approximately 25,500 genes (Arabidopsis Genome Initiative, 2000), but given the possibilities for alternative splicing and polyadenylation during transcript maturation, the transcriptional space of Arabidopsis should be substantially larger. This inference was supported by SAGE results identifying more than 19,000 unique tags in our sampling of a mere 60,000 tags from root tissues alone. Just as was seen in a similarly sized SAGE study of rat fibroblast cells (Madden et al., 1997), the discovery of new tags in the Arabidopsis SAGE libraries had not yet begun to level off at this depth of sampling, and estimates suggested an eventual transcriptome size in the range of 35,000 to 40,000 for the root tissues. However, as noted by Velculescu et al. (1999), to obtain a reasonably complete sampling of a somewhat larger transcriptome (approximately 56,000) using SAGE, one would need to sequence on the order of approximately 650,000 tags. Thus, our data likely represent a low estimate of transcriptome complexity in Arabidopsis.

The size of this SAGE study was similar to a hypothetical situation (15,720 genes; 62,178 sampled tags) modeled by Stöllberg et al. (2000) to estimate the maximum likelihood of generating accurate transcript number and transcript copy frequency given different levels of randomness and non-uniqueness in the transcriptome. Modeling suggested that in a study of this size 1.5% to 6% of SAGE tags should represent two or more genes depending on the non-randomness of sequences represented in the transcriptome. This value matched closely with the 5% figure noted for the 1,045 most abundant tags in this study. However, the modeling study did not take into account the further complexity that might be introduced by variations in transcript maturation processes, such as alternative splicing. Adding further uncertainty to estimates of the Arabidopsis root transcriptome size was the observation that among tags categorized as having ambiguous origin, several could only be matched to antisense sequences for transcripts having ESTs on deposit in GenBank (e.g. CACTTGGATT, the 345th most abundant tag, and TCCGAATCAA, the 385th most abundant tag). SAGE previously demonstrated that antisense transcripts for certain genes were abundant in Caenorhabditis elegans (Jones et al., 2001) and Plasmodium falciparum (Patankar et al., 2001), and although not verified in this study, SAGE tags from antisense transcripts have been verified by reverse transcriptase (RT)-PCR in stressed Arabidopsis plants (G. May, personal communication). Naturally occurring anti-sense transcripts have been seen previously in plants and animals (Terryn and Rouzé, 2000), but it is unclear whether the transcripts identified in our study were somehow involved in posttranscriptional regulation of gene expression via RNAi (Vaistij et al., 2002), or perhaps represented non-coding RNAs that serve some other function in plants (MacIntosh et al., 2001).

Multiphase Mechanisms of TNT Detoxification in Arabidopsis

Studies of the mechanisms plants use to metabolize herbicides and other xenobiotics have previously pointed to a multiphase process for detoxification (Ishikawa, 1992; Kreuz et al., 1996; Ishikawa et al., 1997). Following entrance of the compound into the cell, transformation reactions introduce chemical substituents amenable to conjugation; for example, hydroxyl groups added via enzyme-catalyzed oxidation reactions involving cytochrome P450 enzymes and other mixed function oxidases. The modified compound is subsequently conjugated to GSH or any of several six-carbon sugars through the action of GSTs or UDP:glucosyltransferases, respectively. In the third phase of the process, conjugates are sequestered in the vacuole, where they may be inactivated by further modifications or enter into degradation pathways, or are secreted into the apoplasm where they may be covalently coupled into the cell wall.

SAGE analysis indicates that this multiphase process also functions in the metabolism of TNT by Arabidopsis. Although this has been suggested previously for two other plants, Catharanthus roseus and Myriophyllum aquaticum, based on mass-balance studies of TNT disappearance and metabolite production in axenic cultures (Bhadra et al., 1999a, 1999b), the enzymes most likely involved in the process have not previously been determined. The apparent induction of several cytochrome P450 genes by TNT exposure supports oxidation as the initial transformation step in TNT metabolism. Oxidative transformation of TNT in plants has been suggested from analyses of metabolic products (Bhadra et al., 1999b), but reductive metabolites, such as 2-amino-4,6-dinitrotoluene or 4-amino-2,6-dinitrotoluene, have been recovered most often in mass-balance studies of plant cultures exposed to ¹⁴C-TNT (Lucero et al., 1999; Pavlostathis and Jackson, 1999). Typically, however, the fraction of TNT derivatives recoverable from such studies as small molecules has been only a small percentage of the added TNT, and most of the recovered isotopic label remained in unidentified conjugates or bound to high M_r materials. This suggests that after oxidation of TNT, subsequent conjugation reactions occur too quickly to allow for routine measure of the oxidized intermediates.

While hydroxylation is an effective means of modifying xenobiotic compounds for subsequent conjugation, it rarely serves to decrease toxicity of the parent compound and can sometimes make the compound more toxic. In such cases, hydroxylating enzymes must be coordinately expressed with conjugating enzymes that will quickly decrease the toxic nature of the derivatives. The SAGE data suggest that GSTs were the enzymes shouldering primary responsibility for conjugation reactions involving TNT metabolites. Although previous studies have not provided evidence for the conjugation of GSH to TNT, its conjugation to herbicides and other phytotoxins is well known. In fact, detoxification of herbicides by GSH conjugation is used advantageously in agriculture where compounds that elevate GSH and GST levels, known as “safeners,” are applied to crops before herbicide application (Davies and Caseley, 1999). Many safeners are aromatic compounds with nitrogen-containing functional groups, not unlike TNT, and their metabolism in plants has been relatively well studied. Enhanced glucosylation reactions in response to safener treatment have also been noted (Kreuz et al., 1991) and may explain the 5-fold induction by TNT of a SAGE tag representing Suc-UDP glucosyltransferase (TCTCGAACCT, At5g20830). From these observations it would seem that many of the mechanisms known to function in safener metabolism may be directly applicable in understanding the response of Arabidopsis to TNT exposure.

On the other hand, the 7-fold induction of anthranilate N-benzoyltransferase (Table III), an enzyme that catalyzes the first committed step of a phytoalexin biosynthetic pathway and that has been shown to be adept at using hydroxycinnamoyl-CoA esters to modify anthranilate (Yang et al., 1997), may suggest a conjugation pathway for TNT derivatives that has not been previously suspected. Bhadra et al. (1999b) found that oxidation of the methyl group converted nearly 20% of the TNT added to plant cultures to various benzoyl derivatives. Because benzoate:CoA ligase is a member of an enzyme family that includes 4-coumarate:CoA ligase (Beuerle and Pichersky, 2002), the near 3-fold induction of 4-coumarate:CoA ligase seen in the TNT-treated roots may reflect a mechanism for activating oxidized TNT derivatives before anthranilate conjugation.

Conjugation of GSH with cellular toxins is generally thought to confer increased solubility, decreased toxicity, and increased transport competency on the toxin derivatives (Schröder and Collins, 2002). The transport of GSH conjugates across membranes is an ATP-dependent process handled by ABC transporter proteins (Martinoia et al., 1993; Li et al., 1995; Theodoulou, 2000). As noted in Table III, SAGE detected a 5-fold increase in an ABC transporter (At3g53480) upon TNT exposure. This ABC transporter is of a class related to the Brewer's yeast (Saccharomyces cerevisiae) PDR5 gene and as such, has been suggested to have potential function as a toxin-conjugate efflux pump (Mitterbauer and Adam, 2002). Treatment of liquid-grown Arabidopsis seedlings with the nitroaromatic cytotoxin, 1-chloro-2,4-dinitrobenzene, elicited similar increases in transcript levels for three other ABC transporters, including the AtMRP1 (At1g30400) and AtMRP4 (At2g47800), that are thought to handle transport of GSH conjugates and anthocyanins into the vacuole (Tommasini et al., 1997).

In addition to undergoing oxidation reactions, the TNT taken up by plants is also subject to reduction reactions that target the aromatic nitro-groups, as evidenced by the amino, dinitro-derivatives of TNT identified in plant tissues by Bhadra et al. (1999b). SAGE suggests that such reduction reactions may be catalyzed by NADPH-dependent flavoenzymes, such as 12-oxophytodienoate reductase (At1g76680), which was induced 10-fold by TNT treatment. Similar in sequence to the yeast Old Yellow Enzyme, 12-oxophytodienoate reductase is related to the nitrate ester reductases (Schaller and Weiler, 1997), and another member of this family, pentaerythritol tetranitrate reductase, has been shown to increase TNT tolerance when expressed in transgenic tobacco (Nicotiana tabacum) plants (French et al., 1999). Thus, 12-oxophytodienoate reductase may prove a useful target for overexpression experiments aimed at increasing plant tolerance to TNT and other nitroaromatic compounds.

The observation that two tags representing the NITRILASE2 transcript (At3g44300) and a likely splice variant were strongly up-regulated by TNT exposure were initially of some concern because strong induction of this gene was previously noted during leaf senescence (Quirino et al., 1999). Other leaf senescence transcripts that were induced by TNT exposure included those for γ-VACUOLAR PROCESSING ENZYME (At4g32940) and GLYOXALASE II (At1g53580), which were up-regulated 7- and 3-fold, respectively (data not shown). The transcript for a cinnamyl alcohol dehydrogenase (At5g19440) was also induced 5-fold by TNT exposure, but this was not the same as either of the two CAD genes (SAG25/At4g37990 and SAG26/At1g09500) Quirino et al. (1999) found to be up-regulated in senescent leaves. No tags for either of the CAD genes studied by Quirino et al. (1999) were detected in the SAGE study, nor were tags for the transcript encoding the MtN3-like protein (SAG29/At5g13170) that they examined. Finally, SAGE detected only a 1.3-fold induction by TNT of the ribosomal protein L10 transcript (At1g14320) that was shown to be strongly up-regulated during leaf senescence. Thus, although gene expression patterns in senescing Arabidopsis roots have not been characterized to an extent that allows direct comparison of the two tissues, the SAGE data does not appear to indicate a senescence response in TNT-treated roots similar to that seen in senescent leaves.

Gene Repression in Response to TNT

Inference of metabolic function for genes whose expression is induced by stress can be relatively straightforward given sufficient information on the activity of the encoded proteins. However, understanding the down-regulation of other genes under the same conditions of stress is much more difficult because their repression may be the indirect outcome of regulatory shifts necessary to induce protective genes. Among the transcripts most repressed by TNT exposure (Table IV) were three members of the protease inhibitor/seed storage/LTP family whose products (AIR1A and AIR1B) have previously been associated with lateral root formation in response to auxin (Neuteboom et al., 1999) or response to aluminum (pEARLI-1; Richards and Gardner, 1995). These three genes are adjacent to each other on chromosome 4, but no previous evidence suggests their coordinate regulation. Another group of related genes found to be repressed were the major latex protein homologs, including a Betula vI allergen homolog. Although the function of these proteins is currently unknown, they are generally thought to accumulate in vacuoles. Thus, it is possible that their repression reflects changes in vacuole metabolism brought about by sequestration of TNT-conjugates in this organelle. Changes in vacuolar metabolism might also explain the repression of the tonoplast intrinsic protein and membrane intrinsic protein channels noted in Table IV.

This study was undertaken to improve our understanding of the mechanisms involved in plant tolerance and metabolism of xenobiotic compounds, particularly TNT. Results from these experiments suggest the involvement of previously unappreciated enzymes and, at the same time, strengthen some existing theories of how plants cope with toxic compounds. Identification of the genes involved in the metabolism of TNT should provide for focused engineering attempts to create plants better suited to remediation.

MATERIALS AND METHODS

Plant Material, Growth Conditions, and Root Tissue Isolation

Arabidopsis ecotype Columbia seeds (WT-2, Lehle Seeds, Round Rock, TX) were surface-sterilized and placed in sterile Murashige and Skoog liquid medium prepared according to the manufacturer's protocol (Invitrogen, Carlsbad, CA). Plants were grown for 14 d at 25°C under a 16-h photoperiod with constant shaking at 85 rpm in a growth chamber. TNT was obtained from the U.S. Army Center for Environmental Health Research (Fort Detrick, MD). Toxicity was assessed by adding TNT from a stock solution in dimethyl sulfoxide (DMSO) to yield final concentrations of 0, 5, 10, 15, 20, 25, 30, and 40 mg L^–1 in Murashige and Skoog medium, and cultures were run in triplicate. The cultures were returned to the growth chamber for 5 d, during which time the seedlings in each flask were examined for signs of stress (leaf chlorosis and necrosis). A final concentration of 15 mg L^–1 TNT was judged to produce notable stress in the plants without causing death. Root tissues for SAGE library construction were isolated from seedlings grown 14 d in liquid Murashige and Skoog medium and dosed with TNT to a final concentration of 15 mg L^–1. Control tissues were isolated from seedlings grown under the same conditions and dosed with an equivalent volume of DMSO. The seedlings were grown in the presence of TNT or DMSO for 24 h, after which they were submerged briefly in dH₂O to remove excess medium, and excised roots were immediately frozen in liquid nitrogen. To obtain sufficient biomass, five flasks of seedlings (70–80 seeds in 200 mL of medium in a 500-mL flask) received each treatment, and treatments were replicated on three separate occasions. Root tissues were pooled by treatment and stored at –80°C before RNA extraction.

RNA Isolation and cDNA Synthesis

Total RNA from root tissues was extracted using the LiCl precipitation technique of Chang et al. (1993). Poly(A) RNA was isolated from total RNA using Dynabeads oligo-dT₍₂₅₎ magnetic beads (Dynal Biotech, Lake Success, NY) at a ratio of 0.25 mg of total RNA per 250 μL of Dynabeads and following the manufacturer's instructions. Double-stranded cDNA was synthesized from 5 μg of poly(A) RNA using the Superscript Choice cDNA synthesis kit (Invitrogen) and following the manufacturer's protocol, except for the substitution of a 5′-biotin dT₍₁₈₎ primer in the first-strand reaction.

SAGE Library Construction

SAGE libraries were constructed according to the SAGE Detailed Protocol, v1.0c (Velculescu et al., 1997), a brief description of which follows. Biotinylated cDNAs from each tissue sample were bound to streptavidincoated magnetic beads and digested with NlaIII, a restriction enzyme recognizing the four-base sequence, CATG (anchoring enzyme). DNA released by this digestion was washed away, and the beads, with the adherent 3′ ends of each cDNA, were split into two pools. Linkers containing a binding site for BsmF1 (a type II restriction endonuclease—the tagging enzyme), but different sites for PCR primers were ligated to the NlaIII cleavage site at the 5′ ends of the bead-bound cDNA fragments in each pool. Both pools of cDNAs were digested with BsmF1 to release SAGE tags from the beads, after which the pools were combined, and 102-bp linker-flanked ditags were formed by blunt-end ligation. Following amplification of the ditags by PCR, the linkers were removed by NlaIII digestion, and ditags were ligated to form concatemers. The concatemers were subsequently size-fractionated, ligated into the pZero vector (Invitrogen), cloned, and sequenced.

SAGE Data Analysis

Sequence files were compiled and analyzed using the SAGE Software v3.03, provided by Dr. Kenneth Kinzler (Johns Hopkins University, Baltimore). Tags containing linker sequences and repeated ditags were excluded before analysis. Because the library representing TNT-treated roots was sequenced to a slightly greater extent (32,203 tags for TNT treatment versus 31,973 tags for the control), values for the control library tags were normalized before making comparisons of relative gene expression. Ratios were used to compare the relative expression of tags between the two libraries (e.g. TNT/control), and in instances where a particular tag was absent from a library, a value of 1 was substituted to avoid division by zero. Using the SAGE software, Monte Carlo simulations were performed to estimate the statistical significance of any differential expression. The null hypothesis for these analyses was that the abundance, type, and distribution of transcripts were the same in both libraries. Assuming this null hypothesis, the reported P-chance values represent the fraction of simulations that yielded differences equal to or greater than the observed differences. This is the relative probability of obtaining the observed differences due to random variation, as previously detailed by Zhang et al. (1997).

Gene Identification

To identify the genes from which tags were derived, each 10-base tag plus the 4-base NlaIII recognition sequence was first compared against the AGI database of model genes using the Patmatch analysis tool available on The Arabidopsis Information Resource server (http://www.Arabidopsis.org). If the tag was found to match exactly the NlaIII site closest to the 3′ end of a model gene, this identity was accepted for the tag. Tags that could not be found in the model gene database were compared against all Arabidopsis sequences in GenBank using the same Patmatch tool. Exact matches were annotated accordingly. All listed annotations were current as of July 31, 2003.

Quantitative PCR

Total RNA from Arabidopsis plants grown under conditions identical to those used to generate RNA for the SAGE studies was used to independently verify the expression of selected genes by quantitative RT-PCR. mRNA isolated using Dynal Oligo dT₂₅ magnetic beads served as template for single-stranded cDNA synthesis using Taqman Reverse Transcription Reagents (Applied Biosystems, Foster City, CA). Real-time fluorescent detection of RT-PCR products was performed using an ABI Prism 7700 sequence detection system and Sybr Green PCR Master Mix (Applied Biosystems). The following PCR primers were designed for this study using Primer Express v. 1.0 (Applied Biosystems): cytochrome P450 (At3g28740) forward, 5′-TTGATGCCTTTTGGGATTGG-3′, and reverse, 5′-CAAGGTCACTAGCCGTTGAGC-3′; 60S ribosomal protein L23A (At2g39460) forward, 5′-TCCAGACCAAGAAAGTGAACACA-3′, and reverse, 5′-CATAGTCTGGTGTAAGCCTCACGT-3′; and tonoplast intrinsic protein (At2g36830) forward, 5′-GCTTCTCGGCTCCGTCG-3′, and reverse, 5′-GGCACAGCCAAGCCACC-3′. Amplification reactions were carried out according to the manufacturer's specifications as follows: 2 h at 50°C followed by a 10-min activation of the enzyme at 95°C, and 40 subsequent cycles consisting of 95°C for 15 s followed by 60°C for 1 min. All amplification reactions were run using a dilution series of cDNA from either control or TNT-treated tissues using the same PCR master mix. Amplimers were checked for purity and size by gel electrophoresis to ensure that the correct sequence was amplified. Control reactions omitting reverse transcriptase were run for all samples to ensure that genomic DNA contamination did not contribute to the amplified products. The reported values for qPCR measurements were generated from the C_t values according to the following formula:

where m and y were the slope and y intercept, respectively, for the graph of fluorescence intensity at C_t versus template quantity (log 10).