Functions of chloroplastic adenylate kinases in Arabidopsis.

Adenosine monophosphate kinase (AMK; adenylate kinase) catalyses the reversible formation of ADP by the transfer of one phosphate group from ATP to AMP, thus equilibrating adenylates. The Arabidopsis (Arabidopsis thaliana) genome contains 10 genes with an adenylate/cytidylate kinase signature; seven of these are identified as putative adenylate kinases. Encoded proteins of at least two members of this Arabidopsis adenylate kinase gene family are targeted to plastids. However, when the individual genes are disrupted, the phenotypes of both mutants are strikingly different. Although absence of AMK2 causes only 30% reduction of total adenylate kinase activity in leaves, there is loss of chloroplast integrity leading to small, pale-looking plantlets from embryo to seedling development. In contrast, no phenotype for disruption of the second plastid adenylate kinase was found. From this analysis, we conclude that AMK2 is the major activity for equilibration of adenylates and de novo synthesis of ADP in the plastid stroma.

Nucleotides are essential metabolites in a multitude of biochemical and developmental processes. Nucleotide levels depend on the interaction of purine and pyrimidine de novo synthesis and the respective degradation pathways, salvage reactions that recycle nucleosides and free bases, and phosphotransfer reactions that convert mono-and dinucleotides to the triphosphate form (Zrenner et al., 2006). Nucleoside monophosphate (NMP) kinases catalyze the reversible transfer of one phosphate group of a donor nucleotide (ATP) to an acceptor NMP to generate ADP and a nucleoside diphosphate. Whereas the specificity for the phosphate donor is not very strict, the accepting NMP is recognized with specificity (Rudolph et al., 1999). In many organisms ranging from bacteria to animals, adenylate kinase has been shown to phosphorylate the nucleotide AMP to form ADP (Buchanan and Hartmann, 1959;Neuhard and Nygaard, 1987;Christopherson and Szabados, 1997). Thus, adenylate kinase is required for interconversion of adenine nucleotides (ATP 1 AMP 4 2 ADP) and is considered to be a key enzyme activity in energy metabolism, equilibrating adenine nucleotides in vivo (Pradet and Raymond, 1983;Igamberdiev and Kleczkowski, 2006). Therefore, adenylate kinase plays a fundamental role for the adenylate energy charge that is highly important in all organisms for metabolic processes and growth (Atkinson, 1968;Chapman et al., 1971).
Biochemical studies have shown the presence of this enzyme reaction in plants (Bomsel and Pradet, 1968). The bulk activity of adenylate kinase is located in the chloroplast stroma and in the intermembrane space of the mitochondria (Birkenhead et al., 1982;Hampp et al., 1982;Stitt et al., 1982). However, there are few studies analyzing the distribution of nucleotide pools in relation to metabolism or plant development (Wagner and Backer, 1992;Geigenberger and Stitt, 2000;Igamberdiev and Kleczkowski, 2006). On a cellular level, approximately 45% of the adenine nucleotides are located in the plastid stroma, 46% in the cytosol, and 9% in mitochondria of leaves (Stitt et al., 1982(Stitt et al., , 1989Farre et al., 2001). The reaction catalyzed by adenylate kinase is close to equilibrium in the plastid stroma and in the cytosol, but not in mitochondria (Stitt et al., 1982) due to lack of adenylate kinase activity in the mitochondrial matrix.
Database screening revealed that the genome of Arabidopsis (Arabidopsis thaliana) contains 10 genes with an adenylate/cytidylate kinase signature. To date, two of these genes have been further characterized; At5g26667 was identified as UMP/CMP kinase (Zhou et al., 1998), and At2g37350 was described as a plastidic adenylate kinase (Carrari et al., 2005). Analysis of T-DNA insertion mutants disrupted in At2g37350 revealed increased amino acid levels and enhanced root growth without a strong phenotypic impact (Carrari et al., 2005). Here, we present the functional analysis of two further adenylate kinases (At5g47840 and At5g35170). Both nuclear genes encode isoforms that include a plastid targeting sequence.
However, the individual phenotypes of these two amk (adenosine monophosphate kinase) mutants are strikingly different. Disruption of At5g47840 leads to loss of chloroplast integrity, causing a bleached phenotype from early embryo to seedling development. In contrast, no phenotype for the disruption of At5g35170 was found.

Phylogenetic Analysis of Adenylate Kinase Isoforms
For phylogenetic analysis, we used all 10 Arabidopsis kinases that belong to the same class of aspartokinases (Cheek et al., 2002), including one that has already been functionally characterized (Carrari et al., 2005) and the one that has been described as uridylate kinase (Zhou et al., 1998). We integrated the potato (Solanum tuberosum) adenylate kinase (Regierer et al., 2002) and all members of the same class of rice (Oryza sativa). A phylogenetic tree was constructed containing the complete sequences of the predicted proteins ( Fig. 1A). Several software tools for prediction of intracellular localization were used. Plants stably expressing GFP-fusions with AMK1, AMK2, AMK3, AMK5, and AMK7 were investigated for the subcellular localization of proteins in this family. The phylogenetic tree shows that putative AMKs form distinct branches, whereas all uridylate kinases are on a separate branch (Fig. 1A, blue branch; uridylate kinase activity data not shown).
The first branch of adenylate kinases contains two isoforms of Arabidopsis and two of rice. All these proteins are presumably cytosolic (Fig. 1A, yellow branch), as they are predicted not to have any signal, mitochondrial targeting, or chloroplast transit peptide. This was confirmed by stably transformed GFP-fusion with AMK3 (Fig. 1B, 1 and 2), but in subcellular proteome analyses both proteins (AMK3 and AMK4) were found in mitochondria (Millar et al., 2001;Heazlewood et al., 2004). It has been shown for other cytosolic proteins that association to mitochondria is possible (Giege et al., 2003), but AMK3 and AMK4 without GFP-fusion indeed may localize between the outer and inner membrane of mitochondria to form the previously reported mitochondrial adenylate kinase activity (Birkenhead et al., 1982;Hampp et al., 1982;Stitt et al., 1982).
The more distantly related sequences are the previously characterized Arabidopsis AMK1, the potato ADK, and two further Arabidopsis AMKs together with three homologs of rice (Fig. 1A, gray branch). In comparison with adenylate kinases of other organisms (data not shown), this branch contains exclusively plant sequences. The closest homologs are from endoparasites like Plasmodium falciparum or Babesia bovis. Our study using a stably transformed GFP-fusion with AMK1 and transient expression of a GFP-fusion with AMK7 revealed mitochondrial localization of both proteins (Fig. 1B,(7)(8)(9)(10)(11). In contrast to these findings, there is a previous report describing particle bombardment with an AMK1 GFP-fusion construct and reduced adenylate kinase activity in isolated chloroplasts of mutants with disruption of AMK1. This led to the conclusion that AMK1 is located in plastids (Carrari et al., 2005). These conflicting results of AMK1 localization might be caused by dual targeting, as it has been found for aminoacyl-tRNA synthetases (Duchene et al., 2005;Millar et al., 2006). Further analysis of this detail is not within the scope of this study. Because expression analysis using quantitative real-time reverse transcription (RT)-PCR revealed low transcript levels for AMK6 and AMK7, it is not surprising that these proteins together with AMK1 were not monitored in subcellular proteome studies. Several tools for predicting the subcellular localization resulted in highly probable mitochondrial targeting of most members of this clade.

Correlation Analysis of Adenylate Kinase Expression in Arabidopsis
To elucidate putative specific functions, we screened available expression data for transcriptional co-regulation with all AMKs. Unfortunately, not all genes are represented on the Affymetrix ATH1 microarray including AMK6. Therefore, we also analyzed the expression patterns using quantitative real-time RT-PCR ( Fig. 2A), confirming that AMK1 to AMK5 are much higher expressed than AMK6, while AMK7 is at the detection limit. Using the Arabidopsis co-response database (Steinhauser et al., 2004), we searched for transcriptional correlations using transcript profiles of all Arabidopsis tissues of the AtGenExpress developmental series (Schmid et al., 2005) and the nonparametric Spearman's Rho rank correlation with Bonferroni correction. To visualize significant correlations with genes involved in metabolic pathways, we used the Map-Man ontology, assigning genes to more than 800 hierarchical categories Usadel et al., 2005). Together with the visualization tool PageMan (Usadel et al., 2006) and the statistical analysis using Wilcoxon's test, including the correction of multiple testing by Bonferroni, we identified those categories whose members show a coordinated response together with the respective AMK, which is statistically significant compared with all other genes. P values were assigned a positive or negative value, depending on whether the correlation was positive or negative, and visualized on a false color scale, with increasing blue and red indicating an increasingly and decreasingly significant correlation, respectively. All significant MapMan categories of metabolic processes that match our high stringent selection criteria are visualized in Figure 2B.
The co-response analysis showed distinct expression patterns for AMKs that are located on the previously identified phylogenetic branches. While AMK2 and AMK5 expression was positively correlated with the expression of genes involved in photosynthesis and major carbohydrate metabolism, the expression of AMK3, AMK4, and AMK1 was negatively correlated with genes of photosynthesis and uncorrelated with genes of carbohydrate metabolism. Whereas no significant correlation was detectable between the expression of AMK2, AMK5, AMK3, and AMK4 with genes of glycolysis, tricarboxylic acid cycle, mitochondrial electron transport, and ATP synthesis, a significant positive correlation was found for the expression of AMK1 with genes of light-independent energy metabolism. All these correlations were highly significant. Both genes of the unequivocal plastid branch Predictions of intracellular localization of mature proteins were performed with iPSORT, TargetP, and Predotar. Method intern scores for the predictions are shown, and the results are classified as follows: 0, predicted as having no signal, mitochondrial targeting, or chloroplast transit peptide; M, predicted as having a mitochondrial targeting peptide; P, predicted as having a chloroplast targeting peptide. Our GFP results and the summary of other studies are classified as follows: C, protein found in the cytosol; M, protein associated to mitochondria; P, protein associated to plastids. 1, Millar et al. (2001); 2, Heazlewood et al. (2004); 3, Peltier et al. (2006);4, Peltier et al. (2004); 5, Carrari et al. (2005); 6, Regierer et al. (2002); *, annotated Os08g0288200 is presumably not full length; therefore, an assembly of CT842017 and AK070372 was translated and used for analysis. B, Fluorescence signals of leaf mesophyll cells. 1, AMK3-GFP, GFP fluorescence signal (green); 2, AMK3-GFP, overlay of GFP and chlorophyll autofluorescence signal (red); 3, AMK2-GFP, GFP fluorescence signal; 4, AMK2-GFP, overlay of GFP and chlorophyll autofluorescence signal; 5, AMK5-GFP, GFP fluorescence signal; 6, AMK5-GFP, overlay of GFP and chlorophyll autofluorescence signal; 7, AMK1-GFP, GFP fluorescence signal; 8, AMK1-GFP, DsRed fluorescence signal of the transiently transformed mitochondrial marker DHODH-DsRed (blue); 9, AMK1-GFP, overlay of GFP, DsRed, and chlorophyll autofluorescence signal; 10, AMK7-GFP, GFP fluorescence signal of transiently transformed protoplasts; 11, AMK7-GFP, overlay of GFP and chlorophyll autofluorescence signal; white bars 5 4 mm.
(AMK2 and AMK5) showed identical expression correlations. Both genes of the probable cytosol branch (AMK3 and AMK4) also showed identical expression correlations; however, expression correlations of the plastid and the probable cytosol branch were obviously significantly different. This may lead to the speculation that adenylate kinases are redundant in their respective compartment. Interestingly, no correlated expressions were found with genes involved in nucleotide biosynthesis, although expression of genes involved in pyrimidine and purine de novo synthesis are coregulated to some extent (Giermann et al., 2002;van der Graaff et al., 2004).
Based on phylogenetic studies and the expression co-response analysis, we focused on the functional characterization of highly expressed plastidic adenylate kinases. For this purpose, we used two independent T-DNA insertion lines for the characterization of AMK2 and one line for the characterization of AMK5 (Fig. 3A). The structure of genes and regions of homology between both sequences demonstrate that the C terminus of AMK5 extends that of AMK2 by 290 amino acids. To date, no indication has been found of a possible function of this C-terminal domain.

Isolation of Mutants
Plants with insertions in AMK5 (SALK_000200) or AMK2 (SALK_031816, GABI-KAT_034C05) were obtained from the Salk collection (Alonso et al., 2003) and the GABI-KAT services (Rosso et al., 2003). Using combinations of gene-specific and T-DNA-specific oligonucleotides, the presence of T-DNA inserts was confirmed using PCR (data not shown). Heterozygous T-DNA insertions in AMK2 were identified from SALK_ 031816 (Amk2-1) and GABI-KAT_034C05 (Amk2-2) that did not show differences from wild types. Developing siliques of heterozygous parents contained approximately 25% pale and approximately 75% green seeds (Fig. 3B). Embryos dissected out of these seeds had a pale green to white color (Fig. 3B). Both putative amk2 mutants showed the same whitish seed phenotype. The seed coat was separated, and genotyping of the embryonic tissue confirmed that the whitish embryos are indeed homozygous amk2 mutants, whereas the green embryos segregate into heterozygous mutants and wild types (Fig. 3C). Interallelic crosses of heterozygous amk2-1 with heterozygous amk2-2 plants demonstrated that the white embryo segregation is Rosette leaves of 21-d-old plants on soil, separate roots, or complete 16-d-old seedlings on plates containing one-half-strength MS with 0.5% Suc were analyzed by quantitative real-time RT-PCR. Values are expressed as difference of the ct value to EF1a taken to the power of efficiency. Each bar represents the mean values 6 SD of three individual seedlings; measurements were repeated twice. B, Co-response analysis of AMK genes. The Arabidopsis co-response database was searched for transcriptional correlations using the nonparametric Spearman's Rho rank correlation and Bonferroni correction with the complete AtGenExpress developmental series in wildtype plants (http://csbdb.mpimp-golm.mpg. de; atge0100: developmental series [only WT]). Visualization was done with MapMan (http://gabi.rzpd.de/projects/MapMan) together with PageMan (http://mapman.mpimp-golm. mpg.de/pageman) and statistical analysis of Wilcoxon's test, including the correction of multiple testing by Bonferroni. P values were assigned a positive or negative value, depending on whether the correlation was positive or negative, with increasing blue and red indicating an increasingly and decreasingly significant correlation, respectively. All significant MapMan categories of metabolic processes that match our high stringent selection are shown.

Analysis of Plastid Adenylate Kinases
Plant Physiol. Vol. 146, 2008 maintained in F 1 seed populations (green:white seeds 3:1; data not shown). The fact that the two independent amk2 alleles behave identically and the segregation of the phenotype prove that the observed phenotype is caused by disruptions of AMK2.
Seeds from heterozygous plants that contain homozygous amk2 mutant embryos develop into slightly wrinkled seeds (data not shown). These seeds germinate and grow heterotrophically on medium containing Suc. The rescued homozygous amk2 seedlings were bleached and strongly retarded in growth compared to heterozygous mutants or wild types (Fig. 4A). Under photoautotrophic conditions, both allelic amk2 mutations were lethal. For all experiments, the offspring of heterozygous Amk2 plants was used.
Homozygous amk5 mutants were isolated from a selfed population of SALK_000200. In comparison to wild types, no obvious growth or developmental differences were observed (Fig. 4B).
Transcripts of AMK2 and AMK5 Are Absent in the Respective T-DNA Insertion Lines To determine whether the T-DNA insertion reduces the respective transcript level, quantitative real-time RT-PCR was performed. As shown in Figure 5, virtually no AMK2 or AMK5 transcripts were detected in the respective mutant seedlings, thus confirming that amk2 and amk5 mutants can be regarded as complete knockout for the respective gene product.

AMK Activity Is Reduced in amk2 But Not in amk5
To establish whether the absence of either AMK2 or AMK5 expression results in altered protein abundance, polyclonal antibodies were raised against AMK2. Although the amino acid sequences in the N-terminal regions of the predicted mature proteins are highly homologous, the antibody turned out to be specific for AMK2 and does not cross react with AMK5 protein overexpressed in Escherichia coli (data not shown). Rosette leaves of amk2 plants contain no AMK2 protein (Fig. 6). Most striking is the increase of AMK2 protein in the amk5 mutants, confirming the doubling of AMK2 expression.
To provide evidence that AMK activities have been altered in the mutant lines, we measured total AMK activity of rosette leaves in both directions in heterotrophically grown amk2, amk5, and congenic wild types. Total adenylate kinase activity in amk2 was reduced by 21% in the forward reaction and 37% in the backward reaction compared to wild types, when activity was calculated on total leaf protein basis (Table I). When calculated on a fresh weight (FW) basis, the reduction was a little more pronounced due to an altered protein to FW ratio in amk2 (data not shown). Wild-type activities measured in both directions in rosette leaves of heterotrophically grown Arabidopsis seedlings were approximately 4.1 nkat (mg protein) 21 and in agree-ment with previously reported values of tobacco (Nicotiana tabacum) leaf extracts (8.8 nkat [mg protein] 21 ; Schlattner et al., 1996), but were more than 100 times higher than those reported for Arabidopsis by Carrari et al. (2005). Under any of the tested conditions, total extractable adenylate kinase activity measured in both directions was not changed in amk5 mutants.

AMK5 Encodes AMK Activity
As the complete loss of AMK5 transcript does not result in any reduction of AMK activity, it was tested whether AMK5 really encodes adenylate kinase. Fulllength proteins of AMK2 and AMK5 using pET-DEST42 constructs were overexpressed in E. coli, and protein induction was checked with His-tag specific antibodies (Fig. 7A). Production of AMK2 protein was always higher than AMK5. Purification of the His-tagged proteins proved to be impossible, as binding to nickelnitrilotriacetic acid agarose could not be achieved under native conditions. Therefore, activity measurements   were carried out with crude E. coli extracts together with control inductions. Total adenylate kinase activity of crude E. coli extracts was always higher in lines overexpressing AMK2 than in lines overexpressing AMK5, matching recombinant protein abundance (Fig. 7B). Independent of the variance in protein induction, there was always higher total AMK activity detectable in extracts that produced AMK2 or AMK5 compared to controls. This unequivocally demonstrates that both AMK2 and AMK5 encode adenylate kinase activity.

Adenine Nucleotide Pool Sizes in the Respective T-DNA Insertion Lines
Nucleotide levels were investigated in rosette leaves of heterotrophically grown amk2, amk5, and congenic wild types. In comparison to controls, homozygous amk2 seedlings showed a significant reduction of total adenine nucleotides from 89.5 to 32.8 nmol (g FW) 21 (Table I). Inspection of the individual adenine nucleotides revealed that there was a significant decrease of ATP and ADP and no change of AMP. Calculation of the adenylate energy charge showed a significant decrease in amk2 plants. There is no change in nucleotide levels in plants in the absence of AMK5.

Chloroplast Ultrastructure of amk2 Mutant Lines
The co-response analyses showed that AMK2 transcription is coregulated with genes involved in photosynthesis, and the visual phenotype of amk2 mutants resembles that of mutants disrupted in genes involved in photosynthesis or ''auxiliary'' processes (Maiwald et al., 2003;Bosco et al., 2004;Reiser et al., 2004). Therefore, the total chlorophyll content was measured, and imaging PAM (noninvasive chlorophyll fluorescence measurements) was used to analyze the primary photosynthetic processes in heterotrophically grown amk2 and congenic wild types. As already obvious from Figure 3, the total chlorophyll content of amk2 was strongly reduced (Table II). The quantum yield of PSII (F PSII ) was very low, while the nonphotochemical quenching (qN) was strongly increased in amk2 mutants. F PSII is an empirical index for the assessment of the effective PSII quantum yield, and qN is a complex phenomenon, summarizing components of state transition, photoinhibition, and energy quenching that reflects energy dissipated as heat related to the energetization of the thylakoid membrane.
We used antibodies raised against ATP synthase subunits atpF (subunit I of CF O ) and atpA (a-subunit of CF 1 ) to establish whether the ATP synthase complex is present in chloroplasts of amk2 mutants. Both protein complex subunits were detectable in wild types and in heterozygous mutants, while homozygous Table I. Activity and nucleotide content in amk mutants Rosette leaves of seedlings were analyzed after 21 d of growth on plates containing one-half-strength MS with 0.5% Suc. Activity data are given as percentage of the controls, with forward reaction at 4.2 6 1.3 nmol s 21 (mg protein) 21 and backward reaction at 4.0 6 1.5 nmol s 21 (mg protein) 21 . Each data point represents the mean and SD of seven replicates for amk2, nine replicates for congenic wild-type and heterozygous plants, and four replicates for amk5. Significant differences (P , 0.1) using unpaired twotailed t tests are marked with an asterisk.  amk2 mutants lacked significant amounts of these two subunits (Fig. 8A).
In addition, electron micrographs of wild-type, heterozygous, and homozygous amk2 seedlings were prepared. Chloroplasts of heterozygous and wildtype seedlings showed typical sun-leaf morphology (Fig. 8B, 1). In contrast, chloroplasts of homozygous amk2 mutants had generally lost the integrity of their thylakoid membrane structures and were very heterogeneous in their architecture. Many chloroplasts contained swollen grana thylakoids, whereas stroma thylakoids were rarely observed (Fig. 8B, 2 and 3).
Very few examples of more intact chloroplasts could be found in amk2 (Fig. 8B, 4). Most chloroplasts from homozygous amk2 mutants were estimated to be 50% smaller than in heterozygous or wild-type plants. In addition, chloroplasts of homozygous amk2 seedlings regularly showed an increased number of lipophilic plastoglobuli.

amk2 x amk5 Double Mutants
To further investigate the function of AMK5, we created double mutants by intercrossing heterozygous amk2 and homozygous amk5 plants. Both genes are located on chromosome 5 with a genetic distance of approximately 27 cM between their loci. We determined the genotype of the F 3 offspring from the F 2 amk5 x AMK2-1/amk2-1 and amk5 x AMK2-2/amk2-2 plants. Both populations (209 and 208 seedlings, respectively) contained about 20% seedlings with a whitish amk2-like phenotype. PCR-based genotyping confirmed a Mendelian segregation for the amk2 mutation (data not shown). The slight reduction from the expected 25% homozygous amk2 mutants in the amk5 background can be attributed to the severity of the amk2 mutant phenotype and the weak genetic linkage of AMK2 and AMK5. We did not find additional phenotypic characteristics in double mutants. Thus, no hypothesis of the putative function of AMK5 can be given yet.

DISCUSSION
There are only a few reports of the in planta function of specific adenylate kinase isoforms so far. Regierer et al. (2002) showed that the tuber-specific reduction of plastid adenylate kinase (StpADK) leads to an increase in the adenylate pool and increased starch content in potato tubers. A knockout Arabidopsis mutant defective in AMK1, the closest homolog to StpADK, had a contrasting phenotype, with decreased adenylates, elevated amino acid levels, and reduced starch content in roots but not in leaves (Carrari et al., 2005). In this study, we have classified the entire adenylate kinase gene family using phylogenetic analysis and gene expression correlations. Based on these findings, we addressed two further members of the AMK family that are located in plastids (AMK2 and AMK5).

AMK2 Constitutes an Essential Fraction of Total AMK Activity
Stitt and coworkers found in their study using subcellular fractions of wheat (Triticum aestivum) pro- Table II. Photosynthesis parameter in amk mutants Rosette leaves of seedlings were analyzed after 21 d of growth on plates containing one-half-strength MS with 0.5% Suc. F PSII and qN were determined using imaging PAM. The values were determined for three plants of the same genotype 5 min after the light (250 mmol photon m 22 s 21 ) was switched on. Significant differences (P , 0.1) using unpaired two-tailed t tests are marked with an asterisk.

amk2 AMK
Chlorophyll content (mg g FW 21 ) 13.6 6 5.0* 412 6 83.9 F PSII 0.11 6 0.011* 0.53 6 0.040 qN 0.92 6 0.018* 0.29 6 0.018 Figure 8. Photosynthesis-related analysis of amk2 mutants. A western blot of rosette leaves of 16-d-old seedlings grown on plates containing one-half-strength MS with 0.5% Suc for ATP-synthase subunits. Size-fractionated proteins were immunodecorated with specific antiserum raised against the ATP-synthase subunit proteins as indicated. Lanes 1 and 8, wild types; lane 2, AMK2-1/amk2-1; lanes 3 and 4, amk2-1/amk2-1; lanes 5 and 6, amk2-2/ amk2-2; lane 7, AMK2-2/amk2-2. B, Ultrastructure of chloroplasts from wild-type and amk2 mutant plants. toplasts that 56% of total adenylate kinase activity is in plastids, 3% in the cytosol, and 41% associated with mitochondria (Stitt et al., 1982). Other studies showed that 57% to 72% of total adenylate kinase activity in spinach (Spinacia oleracea) is associated with plastids (Birkenhead et al., 1982). Assuming comparable distribution of adenylate kinase activity in Arabidopsis reduction of only 30% in amk2 mutants would eliminate approximately 38% to 51% of the plastid-associated adenylate kinase activity. Plastid adenylate kinase activity is not exclusively located within the stroma but is also bound to envelope membranes of spinach (Murakami and Strotmann, 1978). Up to now, it was not known whether substantial amounts of plastidassociated adenylate kinase were not inside the stroma of Arabidopsis. Targeting of stably transformed GFPfusions and subcellular proteome analysis (Peltier et al., 2006), together with the severe phenotype of amk2 mutants, demonstrated that AMK2 forms an essential fraction of stromal adenylate kinase. Furthermore, subcellular proteome analysis has found an association of AMK5 with the thylakoid and envelope fraction of plastids (Peltier et al., 2004). Together with the unaltered phenotype of amk5 mutants and the fact that AMK5 cannot substitute the loss of AMK2, it is postulated that AMK5 activity is not located in the chloroplast stroma.

AMK2 Is Important for Chloroplast Integrity
Loss of AMK2 leads to an altered architecture of chloroplasts. This does not prevent embryo development and germination. Similar phenotypes have been observed with mutants in photosynthesis proteins (Maiwald et al., 2003;Bosco et al., 2004), and related phenotypes have been found in purine biosynthesis mutants (Hung et al., 2004;van der Graaff et al., 2004). Functional chloroplasts become crucial after germination when seedlings have to grow independently from resources of the seed storage organs. Mutant amk2 cannot grow autotrophically and show impaired growth even in the presence of sugars and low light. In contrast, mutants in ATP synthase subunits or the Rieske protein are all seedling lethal but are able to survive with exogenous sugar and in low light (Maiwald et al., 2003;Bosco et al., 2004).
It was shown that two important subunits of the ATP synthase are not detectable in amk2. The loss of functional plastid ATP synthase is in agreement with a block of the photosynthetic electron transport chain of amk2, indicated by high qN. This correlates with low abundance of stroma thylakoids, the major location of ATP synthase. Grana thylakoids were present in amk2 mutants but were strongly swollen. From these observations, one might speculate that AMK2 is important for the photosynthetic electron flow by supplying the major part of ADP substrate to the ATP synthase complex. However, it is also likely that the severe chloroplast phenotype is an indirect result of the inability of the chloroplast to equilibrate its adenylates. This is supported by the impossibility to rescue amk2 seedlings by feeding with adenosine or IMP (Supplemental Fig. S2), while mutants in purine biosynthesis could be complemented by feeding nucleotide precursors (Hung et al., 2004;van der Graaff et al., 2004).

Loss of AMK2 Leads to Disruption of the Adenine Nucleotide Pool
Nucleotide measurements showed that amk2 mutants have a reduced adenine nucleotide pool. These results indicate that disruption of AMK2 interferes with de novo synthesis of adenine nucleotides. Furthermore, a decrease of the adenylate energy charge was found. It has to be considered that these measurements include metabolites from plastids, the cytosol, and mitochondria. In mature leaves, approximately one-half of the adenylate pool is present in the plastids, 40% in the cytosol, and the remainder in mitochondria (Stitt et al., 1982). Our data locate AMK2 in the chloroplasts, and other members of the AMK family are presumably responsible for the equilibration of adenylate pools outside the plastid. It is therefore possible that the disturbance of the adenine nucleotides in amk2 is larger in plastids than indicated by our measurements of whole leaf extracts. Subcellular fractionation would be required to confirm this possibility but was not possible due to the small amounts of tissue available from amk2 mutants.
Several observations support the hypothesis that most ATP necessary for plastid-generated metabolites (such as nucleotides, lipids, or amino acids) is produced within the plastid. Most enzymes involved in the purine de novo synthesis are located in plastids (Moffatt and Ashihara, 2002;Boldt and Zrenner, 2003;van der Graaff et al., 2004;Zrenner et al., 2006). Metabolite flux analyses demonstrated that plastids from developing rapeseed (Brassica napus) embryos do not require ATP import to achieve high rates of lipid biosynthesis (Schwender et al., 2004). The identified plastidic nucleotide uniporter is proposed to function as nucleotide exporter of plastids (Leroch et al., 2005), and the two known plastid ATP/ADP translocators presumably supply ATP only to storage plastids (Schü nemann et al., 1993;Neuhaus and Emes, 2000) or are required to import ATP in the dark when the major source of ATP is oxidative phosphorylation in mitochondria (Reinhold et al., 2007). The fact that amk2 embryos fully develop and mature seeds germinate indicates that structural metabolites needed in the early phases of development are likely to be synthesized with the help of cytosolic and mitochondrial ATP pools. During embryo formation and predominantly during embryo germination, salvage processes exceed nucleotide de novo synthesis (Stasolla et al., 2003). While de novo synthesis is largely located in plastids, salvage pathways take place in the cytosol. The severe phenotype of the amk2 mutants indicates that cytosolic nucleotide pools are not able to fully compensate loss of AMK2.

AMK5 Is Not Essential
The complete loss of AMK5 transcript does not result in reduction of total AMK activity or any visible phenotype, indicating that AMK5 is nonessential or redundant with other adenylate kinases. The disruption of AMK5 is alleviated by the doubling of AMK2 expression and increasing AMK2 protein, demonstrating that other AMK activities can compensate loss of AMK5. This is in agreement with the fact that we were able to obtain amk2 x amk5 double mutants with no additional phenotypic characteristics. Both AMK2 and AMK5 show adenylate kinase activity and comparable transcript profiles in young embryos (Spencer et al., 2007) and throughout development. Preliminary results with overexpressed proteins in crude E. coli extracts gave no hint about big differences in substrate specificities (data not shown). In this respect, it is interesting to note that AMK5 has an additional C-terminal domain with no apparent similarity to other proteins or protein patterns. Because subcellular proteome studies have found AMK5 in the thylakoid-and envelopeassociated fractions (Froehlich et al., 2003;Peltier et al., 2004), it is postulated that this protein is not located in the plastid stroma.

CONCLUSION
Adenylate kinases equilibrate adenylates by reversible formation of ADP through transfer of one phosphate group from ATP to AMP. Adenylate kinase activity is encoded by a gene family in Arabidopsis. Functional analysis using T-DNA insertion mutants revealed that only one member of the adenylate kinase gene family is important for equilibration of adenylates and de novo synthesis of ADP in the plastid stroma.
PCR products were inserted into entry vector PENTR/SD/D-TOPO according to the manufacturer's instructions (Invitrogen). For sequence confirmation, positive entry clones were sequenced (MWG Biotech). Clones were recombined into pK7FWG2 plant transformation vector for C-terminal GFPfusion (Karimi et al., 2002) and into pET-DEST42 (Invitrogen) for C-terminal His-fusion using the Gateway LR-Clonase (Invitrogen). Binary vectors were transferred into Arabidopsis plants by Agrobacterium-mediated infiltration (Clough and Bent, 1998). Transgenic plants were selected by kanamycin resistance. Confocal microscopy was performed on 4-to 6-week-old plants grown on soil. For the transient expression of the DHODH-DsRed fusion protein (pUC vector containing 35S promoter-driven expression construct of the first 80 amino acids of dihydroorotate dehydrogenase of tobacco [Nicotiana tabacum] [Giermann et al., 2002] fused to DsRed [CLONTECH]), 5-to 6-weekold leaves were bombarded with the Biolistic PDS-1000/He particle delivery system (Bio-Rad). Bombarded plant leaves were incubated in petri dishes containing wet filter paper, and confocal microscopy was carried out 2 to 3 d later. For transient expression of AMK7 C-terminal GFP-fusion proteins, PEGmediated transfection of Arabidopsis protoplasts was performed (Yoo et al., 2007), and confocal microscopy was carried out 2 d later.

Microscopy
For the localization of GFP-fusions, a confocal microscope system (Leica and Leica Confocal Software LCS) was used. Pictures of dissected siliques and embryos were taken using a SPOT-RT camera (Visitron) attached to an Olympus Systemmicroscope BX41TF (Olympus) controlled by the MetaVue software package (Visitron).
For electron microscopy, pieces of 2 mm 2 from rosette leaves were fixed for 2 h with 2.5% (v/v) glutaraldehyde in 0.1 M sodium-potassium phosphate, pH 7.0, including 2% (w/v) paraformaldehyde and 0.05% (w/v) tannic acid. After washing with phosphate buffer, samples were incubated for 12 h in 50 mM sodium-potassium phosphate, pH 7.0, containing 1% (w/v) osmium tetroxide. After washing with phosphate buffer, the samples were dehydrated in a graded series of ethanol, followed by propylene oxide, incubated in a mixture of propylene oxide/ERL (v/v) and pure ERL (Spurr, 1969), and polymerized overnight at 60°C. Ultrathin sections were contrasted with uranyl acetate and lead citrate. Transmission electron micrographs were obtained with a Siemens 101 at 80 kV. The negatives were scanned with an Epson 1680 Pro Scanner at 1,200 dpi.
All measurements were carried out with plants of the fourth or fifth generation.

RNA Isolation and Transcript Detection
Total RNA was isolated using the NucleoSpin plant kit (Macherey & Nagel), including an on-column DNaseI digestion. For quantitative real-time RT-PCR, complete 16-d-old sterile, culture-grown seedlings were used.
For real-time RT-PCR, single-stranded cDNA synthesis was carried out with total RNA using SuperScript III RNase H 2 reverse transcriptase (Invitrogen). Quantitative two-step RT-PCR was performed using the SYBR Green 1 protocol (Wittwer et al., 1997) and a 7900HT fast real-time PCR system (Applied Biosystems). The oligonucleotides are listed in Supplemental Table S1.

Recombinant Protein Expression in Escherichia coli
Overexpression in E. coli and His-tagged purification of recombinant proteins were done using standard procedures (QIAexpressionist; Qiagen), with the exception that induction was performed at 20°C. Denaturing purification was used to isolate AMK2. Antibodies were produced in rabbits by Eurogentec. Native isolation was applied for activity measurements. As control, pET-DEST42 was used containing a 370-bp fragment of b-lactamase that does not encode any peptide sequence.

Analysis of Enzyme Activities
Frozen plant material was ground in liquid nitrogen using a ball-mill. Then 10-to 20-mg aliquots were extracted by vortexing with 50 to 200 mL of extraction buffer. Composition of extraction buffer was 100 mM Tricine, pH 7.8 (KOH), 10% (v/v) glycerol, 0.1% (v/v) Triton X-100, 5 mM MgCl 2 , 1 mM EDTA, 1 mM EGTA, 1 mM phenylmethylsulfonyl fluoride, 1 mM benzamidine, 1 mM e-aminocaproic acid, and 5 mM dithiothreitol. Extracts were centrifuged at 16,000g, 10 min, 4°C. Measurement was performed in microplates by mixing 2 mL of extract supernatant with 100 mL of measuring buffer following the protocols of Kleczkowski and Randall (1986) for assaying adenylate kinase reaction in direction of ATP and AMP formation (forward) and in direction of ADP production (backward).

Noninvasive Chlorophyll Fluorescence Measurement
Chlorophyll fluorescence measurements were performed using pulse amplitude modulated fluorometer PAM (2000) interfaced with the data acquisition system (Pamwin; Walz). Leaves were dark adapted for 20 min prior to induction fluorescence measurements. Minimal (F o ) and maximal (F m ) fluorescence, F m in the light (F m #), steady-state fluorescence immediately prior to the flash (F t ), and the ratio ([F m # 2 F t ]/F m #) reflecting the F PSII (Maxwell and Johnson, 2000) were recorded at 20°C. qN was determined as 1 2 (F m # 2 F o )/ (F m 2 F o ) (Kooten and Snel, 1990). The intensity of saturating light flashes (800 ms) used for detection of F m # was 4,000 mmol photon m 22 s 21 , and actinic light intensity was 250 mmol photon m 22 s 21 . Chlorophyll content was measured according to Arnon (1949).

Metabolite Analysis
Plant material was quickly frozen in liquid nitrogen and extracts were made using the trichloroacetic acid procedure (Jelitto et al., 1992). Metabolites were measured as done by Schrö der et al. (2005). A spectrophotometric assay was used to quantify AMP (Stitt et al., 1989).

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure S1. Relative transcript levels of nucleotide metabolism genes.
Supplemental Figure S2. Growth of amk2 with adenosine and IMP.
Supplemental Table S1. Oligonucleotides used for real-time RT-PCR.