The specific molecular architecture of plant 3-hydroxy-3-methylglutaryl-CoA lyase

3-Hydroxy-3-methylglutaryl-CoA (HMG-CoA) lyase (HMGL) is involved in branched-chain amino acid catabolism leading to acetyl-CoA production. Here, using bioinformatics analyses and protein sequence alignments, we found that in Arabidopsis thaliana a single gene encodes two HMGL isoforms differing in size (51 kDa, HMGL51 and 46 kDa, HMGL46). Similar to animal HMGLs, both isoforms comprised a C-terminal type 1 peroxisomal retention motif, and HMGL51 contained a mitochondrial leader peptide. We observed that only a shortened HMGL (35 kDa, HMGL35) is conserved across all kingdoms of life. Most notably, all plant HMGLs also contained a specific N-terminal extension (P100) that is located between the N-terminal mitochondrial targeting sequence TP35 and HMGL35 and is absent in bacteria and other eukaryotes. Interestingly, using HMGL enzyme assays, we found that rather than HMGL46, homodimeric recombinant HMGL35 is the active enzyme catalyzing acetyl-CoA and acetoacetate synthesis when incubated with (S)-HMG-CoA. This suggested that the plant-specific P100 peptide may inactivate HMGL according to specific physiological requirements. Therefore, we investigated whether the P100 peptide in HMGL46 alters its activity, possibly by modifying the HMGL46 structure. We found that induced expression of a cytosolic HMGL35 version in A. thaliana delays germination and leads to rapid wilting and chlorosis in mature plants. Our results suggest that in plants, P100-mediated HMGL inactivation outside of peroxisomes or mitochondria is crucial, protecting against potentially cytotoxic effects of HMGL activity while it transits to these organelles.

3-Hydroxy-3-methylglutaryl CoA (HMG-CoA) 3 lyase (HMGL; EC 4.1.3.4) catalyzes the stereospecific cleavage of 3-(S)-HMG-CoA (1). Activity was first detected in pig heart and bovine liver (2,3) where this retro-Claisen reaction leads to formation of acetoacetate and acetyl-CoA. Meanwhile, in vertebrates (4 -12) and bacteria like Tetrahymena pyriformis and Pseudomonas mevalonii (13)(14)(15)(16)(17)(18)(19), HMGL has been functionally, enzymatically, and structurally well-characterized. The enzyme adopts a (␤␣) 8 -barrel fold and belongs to the DRE-TIM metallolyase superfamily (11,19,20). In bacteria, the primary function of this cytoplasmic enzyme is the supply of nutrients allowing cell growth and/or pathogenicity (16,21). Some bacteria also produce acetone as a by-product (22). Particularly in P. mevalonii, HMGL participates in a sequence of catabolic reactions that make the use of mevalonic acid (MVA) as a carbon source possible. In vertebrates, this enzyme is of paramount importance. Genetic disorders impairing HMGL activity in man lead to severe symptoms and ailments, including hypotonia, hypoglycemia, and metabolic acidosis (23)(24)(25)(26). It is therefore not surprising that over the past several years much attention has been focused on the human enzyme (27). It essentially has a dietary function by participating in the catabolism of branched-chain amino acids (BCAAs) through the HMG-CoA cycle and ketogenesis in liver, which is crucial during starvation periods (3,4,(28)(29)(30)(31). The function of HMGL in the catabolism of the BCAA leucine aims to avoid an accumulation of organic acid by-products (25,26). It is initiated by ␤-oxidation of fatty acids that are imported from the blood stream into mitochondria. Generated acetyl-CoA is then converted to HMG-CoA through the action of mitochondrial isoforms of acetoacetyl-CoA thiolase (EC 2.3.1.9) and HMG-CoA synthase (EC 2.3.3.10). Subsequent cleavage of HMG-CoA by HMGL yields acetoacetate, which is used directly or is converted to hydroxybutyrate before exportation again to the blood stream and transport to extrahepatic tissues (29,30).
In contrast, although some studies have indicated that HMGLs are active in some tissues (32)(33)(34)(35)(36)(37)(38)(39), the precise function and subcellular localization of plant HMGLs have yet to be fully characterized. As in vertebrates, plant enzymes participate in the biosynthesis of acetyl-CoA via degradation of the BCAA leucine (39 -42), but there is no clear information available that would assign plant HMGLs with metabolic functions to produce ketone bodies. Thus, alternative functions of HMGL are to be expected. Almost five decades ago, Popják (43) pioneered an innovative concept that envisaged amino acid, fatty acid, and isoprenoid pathways being linked through a common shunt, the so-called MVA shunt. It was confirmed that this shunt operates in vertebrates (44,45), in invertebrates (46), and in higher plants (47,48). Under such conditions, produced acetyl-CoA can be reoriented to fatty acid biosynthesis. Hence, there is some strong evidence that this not well-characterized enzyme HMGL has a larger range of functions than just amino acid catabolism.
To characterize an activity associated with a crude 16,000 ϫ g membrane fraction from etiolated radish seedlings and that interfered with the HMG-CoA reductase (HMGR) test system (34,49), Weber and Bach (50) solubilized under mild conditions and partially purified HMGL by about 154-fold. However, the enzyme was revealed too unstable for complete purification and further characterization. Arabidopsis thaliana contains a unique gene coding for a substantially conserved protein. In a study done by Lu et al. (42) using the Chloroplast 2010 database, this gene had been annotated as coding for an enzyme involved in BCAA degradation (42). The Arabidopsis Information Resource (TAIR) reports on multiple splice variants in Arabidopsis apparently coding for various isoforms (locus AT2G26800). Here, we show that all corresponding proteins possess, as a component, a domain with unknown function specific to plants when compared with functionally identical enzymes from other organisms. Our study aimed to characterize a plant HMGL and its atypical primary protein structure and provide answers for a raison d'être of this plant distinctiveness.

Unusual protein architecture of plant HMGL
Using an in vitro fluorometric assay, it had already been confirmed that the single A. thaliana gene locus 2039548 (At2g26800) encodes an enzyme producing acetyl-CoA (42). Multiple protein isoforms of HMGL have been described in man (51,52). In Arabidopsis as well, the unique gene expresses several transcript variants, annotated At2g26800.1-At2g26800.5, that code for putative HMGL isoforms (https:// apps.araport.org/thalemine/report.do?idϭ69037380&trailϭ 69037380) 4 (75). The molecular masses of these predicted proteins were calculated. Accordingly, a 46,386-Da protein (HMGL46) should be translated from variant 1, a 50,578-Da protein (HMGL51) is produced from variants 2 and 4, a 39,089-Da protein (HMGL39) is translated from variant 3, and finally a 41,393-Da protein (HMGL41) is the product of variant 5. To identify functional HMGL domains, we aligned these putative protein sequences with the well-characterized full-length human HMGL1 (Fig. 1). A highly conserved main core protein (of about 31 kDa) displays strong homologies with the human protein (positions 163-459; Fig. 1). At this stage, it can be hypothesized that this truncated protein corresponds to the active enzyme, which does not seem to be synthesized as such but is rather the product of post-translational proteolytic processing. Alignments show that HMGL39 lacks at the C-terminal part of the protein that play crucial roles in the catalytic activity of the human HMGL (Fig. 1). Thus, only proteins including the C-terminal end "ADASKI" sequence are considered in this study. Moreover, here we focused mainly on two protein variants, HMGL51 and HMGL46, diverging by the length of their nonconserved N termini.
Multiple subcellular localization prediction programs were applied to identify possible targeting functions in the plant HMGL (Table S1). HMGL51 contains a putative mitochondrial transit peptide (35 amino acids; PT35) at its N terminus but also has a putative topogenic signal SKL-like motif (SKI) for import into peroxisomes (PTS1; see Ref. 53) at its C-terminal extremity ( Figs. 1 and 2). HMGL46, the protein product of At2g26800.1, lacks PT35 localized at the N-terminal site of the full-length HMGL51 (Fig. 2). In addition, compared with human HMGL1, HMGL46 comprises a 128-amino-acid N-terminal extension also present in the full-length HMGL51 protein ( Figs. 1 and 2). Further alignments showed a preservation of this extension in all full-length plant HMGL proteins that have been analyzed (Fig. S1). A BLAST data bank search was performed, but no similar sequence or representative of a defined counterpart could be identified. Furthermore, attempts to construct a threedimensional structural model for HMGL46 or HMGL51 failed to identify any homologous sequence, which would give a clue on the function of this peptide (Fig. S2A). In contrast, the part represented by the last 332 amino acids (Figs. S1 and 2), corresponding to a protein with an estimated molecular mass of 34,999 Da (HMGL35), could be easily modeled using the crystal structure of human HMGL as a template (11) (Fig. S2, B and C). From this, it can be concluded that plants synthesize an unusual HMGL protein containing an N-terminal extension with so far unknown functions (P100; Figs. 1 and 2) and the absence of sequence similarities.

Retention of intron 1 favors targeting into mitochondria
An association of Arabidopsis HMGL with uncharacterized organelles has been observed previously (54). To confirm the function of PT35 localized at the N terminus and SKI at the C terminus of HMGL51, we transiently expressed GFP fusion proteins in tobacco BY-2 cells (Fig. 3A). As expected, PT35 offers some gain of function through the import of the reporter protein into mitochondria but not into peroxisomes ( Fig. 2A). In contrast, when the truncated Arabidopsis HMGL35 containing its own C-terminal three-amino-acid motif SKI was fused to the C-terminal end of GFP, the protein was imported into particulate structures clearly corresponding to peroxisomes without any leakage to the cytosol ( Fig. 2A). Along with SKL and SKM, SKI is the most frequent combination found in plant HMGL (Fig. S1B). Deletion of the SKI sequence restored cytosolic localization, confirming that this amino acid sequence is responsible for peroxisomal targeting (Fig. 3A). Interestingly, the deletion also led to nuclear localization (Fig. 3A). To elucidate why HMGL46 lacks these 35 amino acids, nucleotide sequences of HMGL46 (clone AF327420/NM_179756) and HMGL51 (clone NM_179757) were compared with the sequence of the corresponding genomic DNA. This analysis revealed that HMGL51 partially retained the first intron and was most likely the product of alternative splicing. This retention allows the emergence of the start codon ATG necessary for the expression of the full-length HMGL51 (Fig. 3B). In addition, retaining part of intron 1 seems to be responsible for a modification of the secondary structure of the corresponding mRNA that may have a yet unknown role (Fig. S3). We evaluated by RT-PCR relative quantities of each variant in different tissues of Arabidopsis plants (Fig. 3C). Regardless of the variant form, rosette leaves and flowers expressed the gene at rather low levels. Quite in contrast, roots and stems are tissues representative for high expression of both HMGL forms. But in any case, the expression rates of the spliced variant versus the nonspliced variant are rather similar. To confirm this distribution further, we completed a bioinformatics study in which we analyzed 72 sequences (68 EST clones and four full-length cDNAs) that are available in the databases. 26 sequences that allowed an analysis of the sequences at the 5Ј side were collected (Fig. S4). The expression ratio between the alternatively spliced (15) and spliced (11) variants corresponds to 1.3-fold, a value consistent with the results we observed for the RT-PCR evaluation ( Fig.   Figure 1. Sequence alignment of Arabidopsis HMGL51, HMGL46, HMGL41, and HMGL39 with human HMGL1. Identical amino acids are depicted in black boxes, and similar amino acids are in gray boxes. According to Forouhar et al. (19) and Fu et al. (12), catalytic residues and residues that coordinate Mg 2ϩ ion are marked in red. Residues interacting with the CoA ligand are marked with *. The red arrowhead indicates the start of the mature human enzyme without the mitochondrial transit peptide. The underlined helical CoA-binding region displays some mobility upon ligand binding. The underlined glycine-rich loop delineates the signature sequence for the HMG-CoA lyase (HMGCL) family of enzymes. The predicted mitochondrial transit peptide TP35 is highlighted as a yellow segment, the peroxisomal targeting sequence is in red, and P100 is in blue.

Figure 2. Atypical protein organization of plant HMG-CoA lyases. The
A. thaliana full-length HMGL51 protein has been split into the three parts considered in this study. White rectangle, predicted mitochondrial transit peptide (TP35); light gray, additional peptide in plant enzymes with unknown function (P100); dark gray, predicted catalytical entity (HMGL35). SKL is the C-terminal peroxisomal targeting sequence. aa, amino acids. 3C). Altogether, our findings suggest that escaping partial or total splicing of intron 1 favors the targeting of plant HMGL to mitochondria (HMGL51), whereas splicing could possibly drive the enzyme into peroxisomes.

P100's function linked to subcellular compartmentation and control of enzyme activity
To explore the raison d'être for the plant-specific protein architecture, first we tested whether P100 functions as a targeting sequence for specific subcellular localization. GFP fused to P100 was mainly distributed within the cytosol, partially entering the nucleoplasm, but not in peroxisomes or mitochondria (Fig. 4, A and B). When 135 amino acids (PT135 ϭ PT35 ϩ P100) were fused to GFP, the mitochondrial localization was restored without any fluorescence found within the nucleus (Fig. 4, C and D). Then we explored the possibility of P100 altering peroxisomal localization. When GFP was fused to HMGL46, clear peroxisomal localization was impaired because fluorescence appeared in the cytosol and nucleus (Fig. 4, E and F). In contrast, when GFP was inserted between both peptides, peroxisomal localization was restored (Fig. 4, G and H). When deprived of P100, HMGL bearing both targeting sequences was preferentially localized to peroxisomes (Fig. 4, I and J). In contrast, when the fusion protein contained P100 either fused directly to PT35 or to HMGL35, mitochondrial localization became dominant over peroxisomal localization (Fig. 4, K, L, M, and N). However, we were surprised to note that a direct fusion to PT35 displays an almost perfect overlay with mitochondrial localization (Fig. 4, K and L), whereas fusion to HMGL35 led to only a partial localization (Fig. 4, M and N). To exclude that this latter fusion protein is not targeted to a third compartment, a possible localization to the Golgi apparatus, plastids, or lipid bodies was also evaluated (Fig. S5). These possibilities hinged on a function of P100 the putative dominance of one targeting signal over another. P100 interferes with proper recognition of the SKI motif but needs to be fused to HMGL35 in a straight line or needs another peptide in front of it. In addition, it appears that P100 silences a dominant peroxisomal targeting signal versus the mitochondrial targeting signal. Accordingly, Plant HMG-CoA lyase molecular architecture several lines of evidence point to P100 being engaged in proper targeting of Arabidopsis full-length HMGL into mitochondria. Altogether, our results suggest that the full-length HMGL51 is preferentially targeted to mitochondria, not to peroxisomes, and that P100 is essential for this preference.
In a second set of experiments, we tested a possible role of P100 in hampering the catalytic function of HMGL in the cytoplasm. HMGL shares a common substrate with HMGR, the key enzyme in cytoplasmic mevalonate-dependent isoprenoid biosynthesis. Indeed, the production of HMG-CoA is catalyzed by HMG-CoA synthase, and HMG-CoA is used as a substrate by HMGR, which converts it into MVA, necessary for proper cell cycle progression (55). It can be anticipated that if HMGR and HMGL operate simultaneously, their activities might be in competition. Direct interaction of P100 with the catalytic entity could not be demonstrated, but it cannot be excluded that P100 might change the enzyme structure without interacting with it (Fig. S6). Thus, we calculated and compared specific activities from C-terminal His-tagged versions of HMGL35 and HMGL46 using a sensitive radioassay (Fig. 5). Enzymes were partially soluble, but unfortunately, we were not able to purify HMGL46; however, HMGL35 was easily purified using a His tag-trapping system. This observation argues for distinct protein conformations of both enzymes. Consequently, enzyme activities were determined using total soluble protein extracts (Fig. 5A). To avoid some discrepancy due to different HMGL35 and HMGL46 protein concentrations in each extract, a quantification of relative expression and corresponding protein concentration was estimated by Western blotting, and a correction factor was applied for enzyme activity data (Fig. 5B). Although some apparent background activity was detected in Escherichia coli crude extracts (ϳ1000 pmol⅐min⅐mg Ϫ1 ), HMGL35 was about 4-fold more active than HMGL46 (Fig. 5C). Interestingly, although HMGL35 was expressed as a single protein band, HMGL46 appeared unstable, resulting in the detection of four bands (molecular masses, Յ46 kDa) emerging in the bacterial extract (Fig. 5B). One of these bands comigrated with HMGL35. HMGL46 might be degraded by E. coli proteases, and, at least partially, these proteolysis products may be more active than the larger HMGL46. Nevertheless, from this it can be concluded that the shortened HMGL35 version gains improved specific activity as compared with HMGL46. A series of coupled enzyme assays confirmed that the recombinant histidine-tagged protein, HMGL35-His 6 , purified by Ni 2ϩ -affinity chromatography, catalyzes the formation of acetoacetate and acetyl-CoA (Fig. 6). In addition, kinetic properties were evaluated (details can be found in Fig. S7), and a size-exclusion chromatography approach pointed to a homodimeric organization of the active HMGL35 protein. An optimum temperature of 40°C and an apparent activation energy of ⌭ a ϭ 51 kJ mol Ϫ1 was determined from the linearly ascending part of an Arrhenius plot, suggesting that the Arabidopsis enzyme is more efficient than the partially purified radish enzyme with a ⌭ a ϭ 147 kJ mol Ϫ1 (50). The activity of the enzyme depends upon the presence of a divalent cation, preferentially Mg 2ϩ , and shows a pH optimum of 8. No significant positive effect of reducing agents could be observed. Overall, the plant enzyme appears to be less efficient than the bacterial enzyme but more efficient  The formation of acetoacetate was proved by using a coupled assay using HMGL in combination with ␤-hydroxybutyrate dehydrogenase (HBDH) (Sigma, H-9408) in the presence of NADH as described by Gibson et al. (74). The red dot indicates the positioning of the radiolabel. The enzyme reaction mixture was deposited at 2 cm on a cellulose plate, and TLC was developed in butanol/water/formic acid (77:13:10, v/v/v). The migration front is indicated at 20 cm. The chromatograms represent radioactivity registered with a TLC radioscanner (automatic TLC linear analyzer (LB2820-1, Berthold). cts is the number of counts detected in 2 min. Detected substrate (S) and product (P) are indicated. HMGL (10 g) was added to initiate the enzyme assay. The formation of acetyl-CoA (Ac-CoA) was confirmed by using a continuous coupled HMGL-malate dehydrogenase-citrate synthase assay in the presence of NAD ϩ . Specific activities were determined by measuring slopes that allowed us to determine a K m corresponding to 45 M and a V max of 2850 nmol⅐min Ϫ1 ⅐mg Ϫ1 .

Plant HMG-CoA lyase molecular architecture
than the human enzyme (Table S2). However, it must be noted that different assays have been applied, and, moreover, enzyme efficiency (k cat /K m ) cannot be evaluated from all published data.
Taken together, it can be concluded that, similar to vertebrates, a truncated form of the plant enzyme is functional. It seems that the plant-specific P100 acts to balance activity through at least partially slowing down the catalysis. At the same time, the presence of this domain would be necessary to direct the full-length HMGL51 to mitochondria rather than to peroxisomes. Our results corroborate the assumption that there is an alteration of the tertiary and/or quaternary structure, a modification promoted by the fusion of P100 to HMGL35.

Cytotoxic HMGL35 expression in planta
Our final goal was to address the question of why plants would need to suppress HMGL activity in the cytosol? For this purpose, we transformed Arabidopsis plants with a construct that allows expression of a truncated cytosolic form of HMGL35 under the control of a dexamethasone-inducible promoter. We tested the effect of this expression on seed germination and seedling development but also on mature plants (Fig.  7). Induction of active cytosolic HMGL35 expression prevented the correct development of seedlings in germination assays (Fig. 7A). Similarly, after 5 days of induction, mature plants developed symptoms such as wilting and bleached rosette leaves (Fig. 7B). Phenotypic severity depends upon the quantity of protein produced by the transformed line (Fig. 7C). Moreover, it appeared that a threshold expression level is required to result in such phenotypes. Indeed, Arabidopsis line 2 (H-2), which exhibited no clear phenotype, produced low HMGL35 levels as compared with lines H-1 and H-3 (Fig. 7B). From this, it can be concluded that cytosolic expression of an active HMGL in Arabidopsis compromises proper plant development. To gain insight into the possible mechanism induced by HMGL35 expression, chemical complementation with MVA or leucine was carried out, but neither could rescue germination and seedling development (Fig. 8A). To test whether toxicity is plant-specific, overexpression of HMGL was repeated in yeast (Saccharomyces cerevisiae), which lacks HMGL naturally (28). The expression of the Arabidopsis HMGL did not interfere with the proliferation of yeast cells, which might be due to high proteolysis rates observed by immunoblotting (Fig. 8B). Indeed, on solid synthetic minimal medium supplemented with Hollenberg supplement mixture minus uracil (100 mg/liter) and growing on either glucose or galactose (2%), the development of yeast colonies was similar to that of an untransformed control strain. Western blot analysis, where total proteins (50 g) were immunoblotted with the Arabidopsis HMGL35 antibody, confirmed the expression of recombinant proteins. As a positive control, we used purified HMGL35, and the absence of protein in yeast was confirmed by using a protein extract isolated from ScINV1 cells transformed with pYES2. We noted a strong proteolysis rate for both isoforms, as revealed by the presence of small-molecular-mass bands (Fig. 8C). This, at first sight unexpected observation, might be interpreted to mean that yeast cells can somehow escape a compromised MVA pathway by a protective mechanism and thereby avoid a decrease in ergosterol formation needed for growth.
Thus, these results suggest that the toxicity is rather plantspecific, as is the presence of the P100 peptide. Overall, we demonstrated that compartmentation of enzymes using HMG-CoA as a substrate is mandatory in plant cells.

Discussion
Occasionally, some plant enzymes comprise extra peptide domains absent in proteins from other organisms. Such modules make them unique and presumably distinctively regulated. It has been proposed that the human HMGL is redox-regulated via a C-terminal cysteinyl residue (56) responsible for oxidative inactivation and disulfide formation (7). We could not observe any significant effect of reducing agents on recombinant AtHMGL35-His 6 activity in vitro, and moreover, plant HMGLs lack such a C-terminal cysteine, demonstrating that this regulatory option has not been preserved in plants. Instead, plant HMGL proteins include an internal extra domain, P100, with so far undetermined functions.  5 mM). B, expression of Arabidopsis HMGL isoforms is not detrimental to S. cerevisiae growth. HMGL35 and HMGL46 were cloned into galactose-inducible vector pYES. Yeast ScINV1 cells were transformed with pYES2-C, pYES2-HMGL35 (35), or pYES2-HMGL46 (46), and protein expression was induced by addition of 2% galactose. C, separation and detection of HMGL isoforms expressed in yeast ScINV1 cells described in B. Equivalent total protein contents were verified on a Coomassie Blue-stained SDS-polyacrylamide gel, and corresponding Western blot analysis was performed in the same conditions as those described in Fig. 5B. Note the strong proteolytic degradation of HMGL expressed in yeast.

Plant HMG-CoA lyase molecular architecture
The first question that arises is whether plant HMGLs are bifunctional or hypothetical moonlighting enzymes (57), as is the case for the Pseudomonas aeruginosa enzyme (58), which also displays a 3-hydroxy-3-isohexenylglutaryl-CoA lyase activity involved in acyclic terpene catabolism. We did not find any evidence for this. In contrast, here we have addressed a function of P100 in transient inactivation of HMGL and at the same time in the discrimination between mitochondrial and peroxisomal targeting. Another question arising is why specifically plants developed such a strategy to keep HMGL inactive during the transport to these organelles? Several specific properties of plants might explain such differences. First, because the biosynthetic capacity of the mature HMGL version to catalyze the formation of acetyl-CoA and acetoacetate is retained, it can be proposed that, in plants, cytosolic HMG-CoA must be preserved as a substrate. This compound is a central metabolite used for various purposes, in particular for the biosynthesis of isoprenoids (34,35). We postulated that plant HMGL activity had to be inactivated to avoid direct substrate competition with HMGR, a key enzyme in the MVA isoprenoid biosynthesis pathway. But this is unlikely to occur because, surprisingly, expression of the active cytosolic HMGL version induces developmental impairment in Arabidopsis but not in yeast where the enzyme is apparently rapidly degraded. Thus, it appears that this cytotoxicity is plant-specific. Instead, in plant cells, this specificity may indicate a physiological deregulation involving, for instance, phytohormones, of which the exact mechanism remains obscure. In contrast to animals, in plants an equilibrium between synthesis and degradation of BCAA occurs (59). In this context, the BCAA degradation pathway must be differently regulated in plants and mammals. According to transcript expression-level studies, light plays a function in hmgl expression (http://bar.utoronto.ca). 4 Besides, there is increasing evidence that in plants BCAA degradation does not systematically involve HMGL activity (60). Arabidopsis T-DNA insertion mutant seeds show increased levels of isoleucine, leucine, and valine, but it has been postulated that only the leucine catabolic pathway depends on HMGL activity (60,61). HMGL gene expression is stimulated following osmotic stress, bacterial infections, and abscisic acid treatments (http://bar.utoronto. ca/efp/cgi-bin/efpWeb.cgi), 4 suggesting that the protein may function during stress responses to environmental stimuli or senescence induction. Apparently, active enzyme in the cytosol induces a senescence-like phenotype. In more pragmatic terms, it cannot be excluded that under particular stress conditions, proteolysis of inactive HMGL within the cytosol could be induced, and the active form could possibly operate as a plant senescence coregulator or even inducer. This is also consistent with the fact that, under prolonged dark conditions, knockout hml1-2 mutant plants display a yellowish and dehydrated phenotype (62), similar to what we observed when AtHMGL35 was overexpressed in the cytosol under standard growth conditions. The function of HMGL seems not only related to a finetuning of HMG-CoA concentrations in different compartments of a plant cell but also as an element involved in subcellular targeting. In animals and man, HMGLs were first identified within the mitochondrial matrix and in peroxisomes, but more recent studies also suggest an association of some isoforms with the cytosol, the endoplasmic reticulum, and even with some unidentified vesicles when myristoylated (51,63,64). Like its human homolog (65), the Arabidopsis HMGL includes two distinct targeting signals: an N-terminal mitochondrial transit peptide and a C-terminal PTS type 1 motif that guides the protein into peroxisomes. We found no evidence for plant enzymes being hosted by any other subcellular compartment, excluding a function of P100 in targeting to specific organelles. Moreover, our results revealed that expression of an active cytosolic HMGL prevents a normal plant development. This implies that the enzyme needs to be targeted to specific organelles before the active enzyme entity is released by proteolysis. In addition, we noticed that a direct covalent fusion of P100 to the active HMGL35 seems not only to decrease enzyme activity but may also cause alterations in the protein structure. Indeed, using Ni 2ϩ -affinity chromatography, we were unable to purify the HMGL46-His 6 protein, suggesting that the C-terminal end is no longer accessible. Additionally, subcellular localization experiments indicated that, in the presence of P100, peroxisomal targeting is less efficient than in its absence, with even a nuclear localization appearing when the protein is larger (Fig. 3, E and F). With a calculated molecular mass close to 75 kDa, the capacity of GFP-HMGL46 to diffuse through the nuclear pore can be explained by the fact that this protein is a monomer, with the dimer reaching the size limit (66).
In conclusion, plants evolved in a particular way to fine-tune concentrations of HMG-CoA in different subcellular compartments. Our study has generated new questions that must be solved in the future. A yeast two-hybrid based screen or immunoprecipitation of protein complexes will help in identifying new partners interacting with HMGL and more specifically with P100. Finally, a structural elucidation of HMGL35, HMGL46, and HMGL51 proteins would confirm our model.

Materials, chemicals, and nucleic acids
Unicellular organisms, chemicals, enzymes, oligonucleotides, and plasmids are listed in supporting Methods S1 and S2.

General techniques
Details for plasmid construction, transformation, expression, and other procedures are described in supporting Methods S3-S7.

Bioinformatics
Techniques used for computational analyses of gene and sequences and phylogenetic reconstruction as well as protein structure homology modeling are described in supporting Methods S8 and S9.

Transient expression of HMGL-GFP fusion proteins
Tobacco BY-2 cells (67) were transiently transformed by biolistic bombardment with pMRC-derived plasmids harboring GFP and part of the gene coding for HMGL simultaneously with pRFP containing reference genes (68,69). To this end, RFP-SKL specific to peroxisomes/glyoxysomes (68), MITO-Plant HMG-CoA lyase molecular architecture RFP specific to mitochondria (this study; see supporting Methods S6), MAN-RFP specific to the Golgi apparatus (68), and CHLORO-RFP specific to plastids (69), were used. Alternatively, Nile red was used to visualize lipid bodies. Images were acquired using either an LSM 510 (68) or an LSM 700 (70) confocal laser-scanning microscope (Zeiss, Jena, Germany). Images were analyzed using corresponding Zeiss software and exported as TIFF files before being processed using Photoshop 5.0 (Adobe Systems, Mountains View, CA) in which background was reduced using brightness and contrast adjustments applied to the full set of images.

Protein quantification, enzyme overexpression, purification, and assays
pET-HMGL35-His 6 and pET-HMGL46-His 6 were constructed, and protein was expressed as described in the supporting methods. For crude extract analysis, cells suspended in a phosphate buffer A (0.2 M K x PO 4 , pH 7.5, 0.35 M sorbitol, 10 mM Na 2 EDTA, and 5 mM MgCl 2 ) were broken up in liquid nitrogen and centrifuged for 15 min at 10,000 ϫ g. For enzyme characterization, His-tagged proteins were purified using nickel spin columns (Qiagen) according to the manufacturer's protocol. HMGL activity was measured essentially as described by Weber and Bach (50). Details are described in the supporting methods.
For Western blot analyses, 45 g of total Arabidopsis protein extracts or 0.5 g of E. coli cell extracts were separated by DS-PAGE and electroblotted onto a polyvinylidene difluoride membrane (Merck Millipore) before being immunostained as described by Hemmerlin and Bach (55) using rabbit polyclonal antibodies (1:2000) that were raised against gel-purified recombinant HMGL35-His 6 (Institut de biologie moléculaire et cellulaire, Strasbourg, France).

HMGL enzyme assay
HMGL activity was routinely determined as described by Weber and Bach (50) in the presence of 30 g of protein and 9.1 M R,S-[3-14 C]HMG-CoA (0.025 Ci ϭ 55,500 dpm; specific activity, 55 mCi/mmol; Amersham Biosciences) in a total volume of 50 l incubated at 30°C. The reaction was stopped by addition of 125 l of 6 M HCl. The incubation time was chosen such that substrate conversion would not exceed 25%. After transfer of the liquid into scintillation vials, the samples were heated to 110°C (overnight). Under these conditions, the CoA esters are cleaved. 3-(R)-HMG-acid (50% of initial radioactivity) remained in the vial as well as all unused 3-(S)-HMG-acid arising from the natural substrate 3-(S)-HMG-CoA (71). Acetyl-CoA is cleaved into free HS-CoA plus acetate; 14 C-labeled acetoacetate formed by the HMGL reaction is converted into acetone. Both acetone and acetate evaporate under these conditions (4). For determination of HMGL activity in transformed E. coli cells, total soluble extracts were prepared from deep-frozen material. The test system was identical to that described for plant extracts, but much lower protein concentrations were chosen as was appropriate to avoid substrate consumption of more than 30%.

Alternative test systems and product analysis
To prove that the enzyme produced acetyl-CoA from (R,S)-HMG-CoA, a coupled optical test was employed as described by Stegink and Coon (1). The HMGL reaction was coupled to citrate synthase, which in the presence of oxaloacetate converts acetyl-CoA generated through the cleavage of unlabeled HMG-CoA into citrate. Oxaloacetate was produced from malate by NAD ϩ -dependent malate dehydrogenase, and the equimolar reduction of NAD ϩ was monitored at 340 nm (72,73). Alternatively, 14 C-labeled acetoacetate arising from HMGL-catalyzed cleavage of (R,S)-[3-14 C]HMG-CoA could be separated from unreacted substrate by TLC on cellulose plates (20 ϫ 20 cm 2 ; Merck) using the solvent system butanol/water/formic acid (77:13:10 by volume) (74). Aliquots of the test solution described in the preceding paragraph were directly spotted onto the cellulose plates and air stream-dried. Separation from unreacted substrate was achieved by migration for Ͼ4 h. The preceding enzyme conversion of acetoacetate into the more stable hydroxybutyrate (74) was not necessary.

Analyses of gene expression by RT-PCR
Total RNA was extracted from flash-frozen plantlets (500 mg fresh weight), reversed transcribed, and amplified following the procedure described previously (69). Actin2 (AK230311; At3g18780) was used as a reference gene. Oligonucleotides are listed in Table S3.