Resurrecting the Ancestral Enzymatic Role of a Modulatory Subunit*♦

In the post-genomic era, functional prediction of genes is largely based on sequence similarity searches, but sometimes the homologues bear different roles because of evolutionary adaptations. For instance, the existence of enzyme and non-enzyme homologues poses a difficult case for function prediction and the extent of this phenomenon is just starting to be surveyed. Different evolutionary paths are theoretically possible for the loss or acquisition of enzyme function. Here we studied the ancestral role of a model non-catalytic modulatory subunit. With a rational approach, we “resurrected” enzymatic activity from that subunit to experimentally prove that it derived from a catalytic ancestor. We show that this protein (L subunit ADP-glucose pyrophosphorylase) evolved to have a regulatory role, losing catalytic residues more than 130 million years ago, but preserving, possibly as a by-product, the substrate site architecture. Inactivation of catalytic subunits could be the consequence of a general evolutionary strategy to explore new regulatory roles in hetero-oligomers.

An attractive model to study the evolution of allosteric control is ADP-glucose pyrophosphorylase (ADP-Glc PPase), 1 a regulatory enzyme for the synthesis of bacterial glycogen and starch in plants. It has evolved to be regulated by different key metabolites depending on the main carbon assimilation pathway of the organism (1,2). The quaternary structure plays an important role in its allosteric regulation. Most of the bacterial ADP-Glc PPases are homotetramers (1), whereas the plant enzymes comprise two types of homologous but distinct subunits (2)(3)(4)(5), the small (S subunit, 50 -54 kDa), and the large (L subunit, 51-60 kDa). They form a heterotetrameric structure S 2 L 2 where the S subunit is catalytic, but L subunit is modulatory without catalytic function (2,6). The L subunit modifies the sensitivity of the S subunit toward allosteric effectors, possibly by protein-protein interactions. This is of remarkable physiological importance because expression of different L subunits in different plants tissues confers distinct allosteric prop-erties to the tetramer according to the necessities for starch synthesis (7). The similarity between S and L subunits (ϳ50% identity) suggests a common ancestry (5). Therefore, gene duplication, divergence leading to new roles (catalytic and modulatory), additional divergence of modulatory genes, and further establishment of differential tissue expression seems to be the strategy used by nature to control and regulate the synthesis of starch, one of the most important storage compounds on earth. To understand the evolutionary path that led to this divergence, it is fundamental to know the ancestral role of these subunits.
The ancestor of S and L subunits could have been 1) catalytic as the S subunit or 2) non-catalytic as the L subunit. The first scenario is supported by the similarity between plant subunits with many active bacterial ADP-Glc PPases (1). However, we cannot simply rule out the second possibility because of the existence of non-catalytic homologues such as the products of the glgD genes from Gram-positive bacteria (8). There is strong evidence that the L subunit from the potato (Solanum tuberosum L.) tuber ADP-Glc PPase binds substrates. The heterotetramer (S 2 L 2 ), as well as bacterial homotetramers (␣ 4 ), binds four ADP-[ 14 C]glucose molecules (9,10). We hypothesized that the L subunit maintained the structure of the substrate site needed for binding, whereas the catalytic ability was severely reduced by mutations of essential residues. To test this hypothesis, we sought to design an L subunit with significant catalytic activity mutating as few residues as possible.

DNA Methods
The Macromolecular Structure, Sequencing, and Synthesis Facility (MS 3 F) at Michigan State University performed the synthesis of oligonucleotides and automated DNA sequencing.

Site-directed Mutagenesis
Mutant D145N on the S subunit was obtained as described previously (6). Combined chain reaction (CCR) was used to introduce other mutations (11). The plasmids encoding the wild-type S subunit (pML10), the S subunit mutant D145N (pML10-D145N), or the wildtype L subunit (pMON17336) were used as templates for the mutagenesis (12,13). In the S subunit, the residues were changed with the following oligonucleotides: R33K, 5Ј-GGA GCT GGG ACC AAA CTT TAT CCT CTA-3Ј; K43T, 5Ј-AAA AAA AGA GCA ACC CCA GCT GTT CCA-3Ј; K198R, 5Ј T GAA TTT GCA GAG CGC CCG CAA GGA GAG C-3Ј. In the L subunit, the oligonucleotides used were: K44R, 5Ј-GGA GAA GGG ACC CGC TTA TTC CCA CTT A-3Ј; T54K, 5Ј-AGT AGA ACT GCA AAG CCT GCT GTT CCG-3Ј; K213R, 5Ј-AG TTT GCT GAA CGT CCA AAA GGT TT-3Ј. L subunit mutants on the residue 160 were obtained as described previously (6). The mutated fragments were exchanged to the double mutant plasmid pMON17336-K44R/T54K using the restriction sites SspI and SacI to obtain the triple mutants L K44R/T54K/D160N and L K44R/T54K/D160E . The coding regions of the mutant plasmids were sequenced to confirm that only the intended mutations were introduced.

Protein Methods
Protein concentration during enzyme purification was measured by using bicinchoninic acid reagent (14) from Pierce, with bovine serum albumin as standard. Interfering substances were previously removed precipitating with 12% trichloroacetic acid and 0.025% sodium deoxycholate and dissolving in 5% SDS, 0.1 N NaOH. Protein concentration of the purified enzymes was determined by UV absorbance at 280 nm using an extinction coefficient of 1.0 (6). Electrophoresis (SDS-PAGE) and immunoblotting of the protein samples were performed as described previously (6). Samples were desalted with Bio-Rad 10 DG chromatography columns. Centricon-30 devices (Amicon Inc.) were used to concentrate the enzymes.

Enzyme Assays
Pyrophosphorolysis Direction-The formation of [ 32 P]ATP from [ 32 P]PP i and ADP-Glc was measured as follows. The standard aqueous reaction mixture contained 50 mM Hepes buffer (pH 8.0), 7 mM MgCl 2 , 2 mM dithiothreitol, 1 mM [ 32 P]PP i (300 -1500 cpm/nmol), 2 mM ADP-Glc, 10 mM NaF, 2 mM 3-PGA, and 0.2 mg/ml bovine serum albumin, plus enzyme in a total volume of 0.25 ml. After 10 min of incubation at 37°C, the reaction was terminated by the addition of 3 ml of cold 5% trichloroacetic acid. The [ 32 P]ATP formed was measured as described previously (3).
Unit Definition-In the above assays, one unit of enzyme activity is equal to 1 mol of product, either [ 32 P]ATP or [ 14 C]ADP-glucose, formed per minute.

Expression and Purification of Mutant Enzymes
The Escherichia coli AC70R1-504 cells, which lack endogenous ADP-Glc PPase activity (12), were co-transformed with compatible plasmid pML10 (S subunit) and pMON17336 (L subunit) or their respective mutated derivatives for co-expression of the subunits. Transformed cells were grown in 2 liter of LB medium, induced, harvested, and sonicated in buffer A (50 mM Hepes (pH 8.0), 5 mM MgCl 2 , 0.1 mM EDTA, 10% sucrose) as described previously (16). The purification was performed at 0 -4°C. The crude extract was applied onto a 12-ml DEAE-Sepharose column (Amersham Biosciences) equilibrated in buffer A, and eluted with a linear NaCl gradient (10 column volumes, 0 -0.5 M). Purest fractions were pooled, desalted, concentrated, and applied onto a Mono Q HR 16/10 (FPLC, Amersham Biosciences) column equilibrated with buffer A and eluted with a linear NaCl gradient (16 bed volumes, 0.1-0.5 M). The post-Mono Q fractions were pooled, concentrated, resuspended in buffer B (buffer A plus 1.3 M ammonium sulfate), and applied onto a phenyl-Superose HR 5/5 (FPLC, Amersham Biosciences) column equilibrated with buffer B. The sample was eluted with a decreasing linear gradient (20 bed volumes) of ammonium sulfate (1.3 to 0 M ammonium sulfate). The purest fractions were pooled, concentrated, and desalted. After this procedure, enzymes were Ͼ95% pure. The enzymes were followed by pyrophosphorolysis activity, except S D145N L WT , S D145N L K44R , and S K43T L WT , which were followed by SDS-PAGE and immunoblotting.

Calculation of Kinetic Constants
Curves were performed varying the concentration of the effector studied while keeping a constant saturated concentration of the other components of the standard reaction mixture. The kinetic data were plotted as initial velocity in nmol min Ϫ1 (v) versus effector concentration in M ([S]). This plot was used to fit the parameters of a modified Hill equation (17), v ϭ V max [S] n /(S 0.5 n ϩ [S] n ), by the Levenberg-Marquardt non-linear least squares algorithm provided by the computer program Origin® 5.0 (18). The parameter S 0.5 is the concentration of substrate needed to obtain 50% of the maximum activity (V max ). The activation curves were fitted with the similar equation The parameter v 0 is the activity in absence of activator, and A 0.5 , is the concentration of activator needed to obtain 50% of the maximal activation (V max Ϫ v 0 ). The standard deviations were obtained by the Levenberg-Marquardt method (18). Specific activity was determined at saturating concentrations of effectors and different dilutions of the enzyme to plot activity versus protein concentration. The specific activity was determined from the slope in the linear range. From this value, k cat was calculated as number of molecules of ADP-Glc synthesized (consumed in the reverse direction) per molecule of enzyme per second. For this calculation, we used 202 kDa as molecular mass of the S 2 L 2 enzyme (12).

Identification of Putative Essential Missing Residues in the L Subunit
The set of invariants from an alignment of catalytic ADP-Glc PPase subunits should contain most of, if not all, the essential residues for catalysis. Among those, the ones that are absent in the L subunits were candidates to be responsible for the lack of catalytic activity. In the comparison, to avoid missing important residues, we excluded sequences of modulatory or potentially inactive ADP-Glc PPase subunits. A first conservative alignment was performed with ADP-Glc PPases with experimentally proven catalytic activity (Fig. 1A). Those enzymes were from Rhodobacter sphaeroides (19), Agrobacterium tumefaciens (20) (23), Hordeum vulgare, endosperm S subunit (24), Arabidopsis thaliana S subunit (7), and S. tuberosum, tuber S subunit (12). In that alignment, 64 residues were invariant in the putative catalytic domain of the potato tuber S subunit (residues 21-290 (2, 25)). From those, the only four absent in the potato tuber L subunit were Arg 33 , Lys 43 , Gln 126 , and Ile 232 (S subunit numeration).
A second comparison with a broader and less strict selection of enzymes strongly suggested that Gln 126 and Ile 232 are not essential. The criteria to select enzymes with high probability of being catalytic in this second round were: 1) from bacterial genome projects that detected only one ADP-Glc PPase gene per organism and 2) genes identified as glgC (active) and not as glgD, known to produce non-catalytic polypeptides in Bacillus (8). Arg 33 and Lys 43 were still invariant in this second alignment, but Gln 126 and Ile 232 were replaced in 7 and 12 out of 58 cases, respectively. Among S subunits from plants, Arg 33 and Lys 43 were conserved except sequence 20 (homologous of Lys 43 to Asn) and sequence 21 (homologous of Arg 33 to Lys) (supplemental Table I). The existence of duplicated genes (paralogs) and pseudo-genes in plants raised the question whether these exceptions actually generate active enzymes.

Structure Prediction and Homology Modeling
Homology modeling of the putative pyrophosphorylase domain (residues 31-310) of the wild-type L subunit of the ADP-Glc PPase from potato tuber and the double mutant L K44R/T54K was performed with the program Modeller 6v2 (26). The structure was modeled using two simultaneous templates; the known atomic coordinates of the E. coli dTDP-Glc PPase (RffH) complexed with dTTP and Mg 2ϩ (Protein Data Bank code 1MC3), and human Agx2 (UDP-N-acetylglucosamine pyrophosphorylase; Protein Data Bank code 1JVG) (27,28). In this way, we took advantage of the features of the program Modeller to use the best part of each template (29). We guided the initial alignment for the homology modeling with the secondary structure prediction performed with the PSI-PRED method (30) available on the PSI-PRED server (bioinf.cs.ucl.ac.uk/psipred/) (31). Alignment with the templates was based on homology, secondary structure, and hydrophobicity.
There were strong indications that ADP-Glc PPases share with the templates a pyrophosphorylase domain with the same fold despite the identity was ϳ20% (2). In addition, threading and fold recognition analysis with the program GenTHREADER (32) predicted a common fold with a very high confidence. Running the region from residues 31-320 from the L subunit from potato tuber, gave E values (expected number of false hits per sequence query) of 2 ϫ 10 Ϫ5 , and 4 ϫ 10 Ϫ5 for 1G0R (dTDP-Glc PPase) and 1JV1 (UDP-N-acetylglucosamine pyrophosphorylase), respectively. In the active site, we modeled a deoxyribose triphosphate ligand extracted from the coordinates of the dTTP bound to RffH. That is the shared moiety between the ATP and dTTP bound to the ADP-Glc PPase and dTTP-Glc PPase, respectively.
The model was checked with the programs Verify3D (33) and PRO-CHECK (34). Minor adjustments in the alignments with the templates were checked and modified iteratively to improve the three-dimensional-one-dimensional scores provided by Verify3D along the sequence. There was no structural change, in any of the iterations, in the area of the residues analyzed in this work. The region between residues 217-232, which corresponds to an insertion, was the only one with low 3D-1D score. This is expected from insertions of more than eight residues (35). However, this area is very far from the region analyzed in the paper, in the other lobe of the catalytic domain, exposed, and not facing the active site.
Recently, the crystal structure of the tetrameric form (S 4 ) of the S subunit from potato tuber was solved (36). This structure was in an inhibited conformation, which made it unsuitable as a template for modeling an active L subunit. However, it confirmed that the templates selected in this study (RffH and Agx2) were appropriate, because they share the same fold in the catalytic domain as previously predicted (2). RffH and Agx2 are not allosterically regulated and are better templates to model the active conformation of the substrate site of the L subunit.

Phylogenetic Analysis
ADP-Glc PPase sequences from Supplemental Table I were obtained from the National Center for Biotechnology Information (www.ncbi. nlm.nih.gov). The expression data were taken from the literature. Multiple sequence alignment was performed by the Clustal program (37) available on-line (clustalw.genome.jp), and afterward, it was manually refined with the BioEdit v6.0.7 program (www.mbio.ncsu.edu/BioEdit/ bioedit.html). For the phylogenetic analysis, the N terminus of all the sequences was discarded. That region is not homologous and corresponds to the leading peptide for subcellular targeting, which has been shown in the S subunits that belongs to an exon with a different evolutionary history (38). Alignment started in residue 2 of the mature recombinant S subunit from potato tuber (A 2 VSDSQNSQT . . . ) and residue 13 of the mature recombinant L subunit from potato tuber (Q 13 TVFVDMPR . . . ). The unrooted tree was built by the neighborjoining method (39) as implemented in the PHYLIP package (evolution-.genetics.washington.edu/phylip.html), version 3.6. The tree was drawn with the program Phylodraw (40). The outgroup chosen was the ADP-Glc PPase from Anabaena because cyanobacterial enzymes are the closest to plant enzymes (2). The topology of the tree of the L subunit gene family from plants was further investigated using the Maximumlikelihood method as implemented in the program TREE-PUZZLE (41). The tree was rooted with the sequence of the L subunit from the unicellular algae Chlamydomonas reinhardtii and plotted with the program TREEVIEW (42).

RESULTS
Identification of critical missing residues was the first important step to design an active L subunit. A sequence alignment of ADP-Glc PPases with reported activity (experimentally assayed) allowed us to focus on few candidates. The set of invariants should contain most of, if not all, the essential residues. The subset of the ones absent in the L subunit was of special interest. We chose Lys 44 and Thr 54 in the L subunit from potato tuber as the best candidates to study because the homologous residues (Arg 33 and Lys 43 in the S subunit, Fig. 1A) were differentially conserved in the active bacterial and S subunits (see a detailed analysis under "Experimental Procedures"). In addition, Lys 44 and Thr 54 are in a highly conserved region of ADP-Glc PPases (Fig. 1A).
Resurrection of ADP-Glc PPase Catalysis-To convert the modulatory L subunit into a catalytic one, we substituted Lys 44 and Thr 54 by Arg 44 and Lys 54 , respectively. The mutant L K44R/ T54K was expressed alone, but no activity was detected, most probably because the L subunit is unable to form a stable tetramer in absence of the S subunit (43). The activity of L subunit mutants cannot be readily tested with a co-expressed wild-type S subunit because of the intrinsic activity of the latter. We co-expressed the L subunit mutants with S D145N , an inactive S subunit in which the catalytic Asp 145 was mutated (6). In this way, the activity deriving from the S subunit was reduced more than three orders of magnitude (Table I). Coexpression of the L subunit double mutant L K44R/T54K with S D145N generated an enzyme with considerable activity, 10 and 18% of the wild-type enzyme (S WT L WT ) in the forward and reverse direction, respectively (Table I). Single mutation K44R generated an enzyme (S D145N L K44R ) with no significant activity over the control (S D145N L WT ). Mutation T54K (S D145N L T54K ) was more effective, but the combination of both mutations in the L subunit (S D145N L K44R/T54K ) had the most dramatic effect (Table I). Therefore, we concluded that the two residues Arg 44 and Lys 54 are needed for restoring catalytic activity to the L subunit.
The resurrection of catalytic activity in the L subunit by incorporation of Arg 44 and Lys 54 predicted that the homologous residues in the S subunit are also critical. Replacement of those two residues by the homologous in the L subunits (mutations R33K and K43T) decreased the activity one and two orders of magnitude, respectively, in both directions, confirming the hypothesis (Table I). The mutant enzymes were still activated by 3-PGA and inhibited by orthophosphate (P i ) (data not shown), which are the main regulators of plant enzymes (2). The wildtype enzyme and S D145N L K44R/T54K had very similar kinetic properties indicating that the architecture of the substrate site has been conserved. The apparent affinities for the substrates in both directions were in the same range and the allosteric properties of S D145N L K44R/T54K resembled those of the wild type as well (Table II). The presence of 2 mM P i decreased the apparent affinity of 3-PGA for both the wild type and S D145N L K44R/T54K indicating that this new form has a similar sensitivity to P i inhibition (Table II).
The fact that only two mutations in the L subunit restored enzyme activity is the ultimate evidence that the L subunit derives from a catalytic ancestor. To confirm that the catalysis occurs in the L subunit of S D145N L K44R/T54K , we disrupted the substrate site in each subunit and compared the kinetic properties. Previously, in the wild-type enzyme, replacement of Lys 198 in the S subunit decreased the substrate affinity (Glc1P), whereas disruption of the homologous residue Lys 213 in the L subunit did not have the same effect (9). In the case of S D145N L K44R/T54K , the mutation K213R on the L subunit severely decreased the apparent affinity for Glc1P, whereas K198R on the S subunit did not (Table III). This indicated that the L subunit double mutant, and not S D145N , was catalytic. In the wild-type enzyme, Lys 213 does not seem to play any important role; but in S D145N L K44R/T54K , it recovered its ancestral ability to confer to the enzyme a high apparent affinity for Glc1P. Previous results showed that Asp 145 in the S WT subunit is essential for catalysis, whereas the homologous Asp 160 in the L WT subunit is not (6). Conversely, in this study, mutation D160N or D160E in the L K44R/T54K subunit abolished the activity (Table IV), which shows the ancestral essential role of this residue and confirms that the catalysis of S D145N L K44R/T54K occurs in the L subunit.
Structural Basis for the Ancestral Function-A comparative model of L K44R/T54K illustrates the predicted role of Arg 44 and Lys 54 (Fig. 1B). In the model, Asp 160 , which is homologous to the catalytic Asp 145 in the S subunit and catalytic Asp 142 in the E. coli ADP-Glc PPase (6,25), interacts with Lys 54 . This type of interaction (Lys 54 -Asp 160 ) has also been observed in crystal structures of enzymes that catalyze similar reactions, such as dTDP-glucose pyrophosphorylase (dTDP-Glc PPase) and UDP-N-acetylglucosamine pyrophosphorylase (UDP-GlcNAc PPase), and postulated to be important for catalysis by correctly orienting the Asp (28,44,45). In addition, Lys 54 interacts with the oxygen bridging the ␣and ␤-phosphates as it has been observed in the crystal structure of E. coli dTDP-Glc PPase (28). That contact may neutralize a negative charge density stabilizing the transition state and making the pyrophosphate a better leaving group. Arg 44 interacts in the model with the ␤and ␥-phosphates of ATP, which correspond to the PP i product (Fig. 1). Similarly, Arg 15 in the E. coli dTDP-Glc PPase was postulated to contribute to the departure of PP i (44). The kinetic data agreed with the predicted interaction of PP i with Arg 44 . A Lys in that position, in both the catalytic L subunit mutant and the S subunit, decreased the apparent affinity for PP i at least 20-fold (Fig. 2). In the model of the wild-type non-catalytic L subunit, Lys 44 and Thr 54 do not interact as Arg 44 and Lys 54 (Fig. 1).
Phylogenetic Analysis-A phylogenetic tree of the ADP-Glc PPases from photosynthetic eukaryotes provides information about the origin of the subunits. In multicellular organisms, S and L subunits form two and four distinct groups, respec- The reaction in the forward (synthesis of ADP-Glc) direction uses ATP and Glc1P as substrates releasing PP i and ADP-Glc as products. The reaction can also be assayed in the reverse (pyrophosphorolysis of ADP-Glc) direction. Enzymes were obtained by co-expression of the S and L subunits and purified to homogeneity, and the activity was assayed in both directions of the reaction as described under "Experimental Procedures." WT, wild type.  tively (Fig. 3). The two main groups of S subunits are from dicot and monocot plants, whereas L subunit groups correlate better with the reported tissue expression. The first L subunit group, generally expressed in photosynthetic tissues (supplemental Table I), comprises L subunits from dicots and monocots. Group II displays a broader expression pattern, whereas groups III and IV are expressed in storage organs (roots, stems, tubers, seeds). Subunits from group III are only from dicot plants, whereas group IV are seed-specific subunits from monocots. These last two groups stem from the same branch of the phylogenetic tree and split before monocot and dicot separation (Fig. 3). DISCUSSION The resurrection of the enzymatic activity of the L subunit ADP-Glc PPase by only two mutations reveals the catalytic ancestry. In addition, it points out the important role of the residues Arg 44 and Lys 54 for catalysis. A comparison with other pyrophosphorylases agrees with this data.
The importance of Arg 44

and Lys 54 in the Active L Subunit Mutant Is Possibly Widespread in Other Pyrophosphorylases-
The identity between ADP-Glc PPases and other nucleotidediphosphate-sugar pyrophosphorylases is very low (ϳ20%). However, prediction of the secondary structure and further structure-based alignment of the sequences emphasized the similarities (1,2,25). Residues Arg 44 and Lys 54 of the active L subunit mutant are not only present in active ADP-Glc PPases but also in other pyrophosphorylases in structurally similar positions (25,44,45). In addition, in some cases these residues were shown to be important. For instance, Lys 25 in the UDP-GlcNAc PPase (GlmU), homologous to Lys 54 in the mutant L K44R/T54K , was mutated to Ala and the specific activity decreased 8-fold (45). In this work, homology modeling predicted that Arg 44 , an important residue for activity in the mutant L K44R/T54K , interacts with PP i making it a better leaving group in the forward (synthesis) direction. In the reverse (pyrophosphorolysis) direction, it is predicted to have a role in PP i binding as a substrate. In other pyrophosphorylases, a non-conservative mutation from Arg to Ala, at the homologous position, reduced the specific activity in the nucleotide-sugar synthesis direction two to three orders of magnitude (45,46). Unfortunately, the effect on the PP i apparent affinity has not been examined. The presence of homologous residues to Arg 44 and Lys 54 suggest that they may be also important in other distantly related pyrophosphorylases, such as the enzyme 4-diphosphocytidyl-2-C-methylerythritol synthetase (47). In this case, despite the very low percentage of identity, the homology can also be identified with a structural alignment.
Evolutionary History of ADP-Glc PPase L Subunits-We cannot ascertain when the L subunit from potato tuber became modulatory only by phylogenetic analysis, which is based upon primary structure rather than function. However, with structure-function relationship information, we could theoretically trace function or the absence of it in the tree. Both groups III and IV lack the residues homologous to Arg 33 and Lys 43 (S WT numeration, Fig. 1A). Group III has Lys, His, and Gln, and group IV has Gln, respectively, in place of a homologous Arg 33 (Fig. 3). A distinctive characteristic in both groups is that a Thr replaces the catalytic Lys 43 (Fig. 3). The most likely scenario supported by phylogenetic analysis (Fig. 3) is that the shared ancestor of both branches lacked a Lys in that position and, consequently, was already non-catalytic. Because one branch (III) is composed of dicot and the other (IV) of monocot plants, we can infer that the L subunit from groups III and IV ceased to be catalytic before monocots split from dicots among the angiosperms more than 130 million years ago (48,49). Conceivably, they had already acquired adaptive modulatory properties at that point in evolution given that catalysis had been already lost (Fig. 3).  Table I. To determine specific activity, concentrations of effectors were saturating. The increase values were the relative values of S 0.5 compared to the enzyme S D145N L K44R/T54K (Table II) 2. Effect of the mutations R33K (S subunit) and K44R (L subunit) on the pyrophosphate saturation curves. Activity was analyzed in the pyrophosphorolysis direction as indicated under "Experimental Procedures" for the wild-type enzyme and the mutants S R33K L WT , S D145N L K44R/T54K , and S D145N L T54K . Complete saturation could not be reached for the mutants S R33K L WT and S D145N L T54K . Higher concentrations of PP i lead to precipitation and cloudiness of the reaction mixture. The kinetic constants were calculated as described under "Experimental Procedures." Modulatory L subunits from branches I and II of the phylogenetic tree may be non-catalytic too (7), but the reasons for this lack of activity are not clear. They contain in their sequence the important residues analyzed in this work (Fig. 3), as well as all the others conserved in the catalytic subunits. This presence predicts a catalytic role; however, the only biochemical evidence gathered so far has not confirmed it. None of the L subunits from A. thaliana (groups I-III, supplemental Table I), when expressed in presence of an inactive S subunit (APS2), originated a form with detectable activity (7). Despite no apparent essential residue seems to be missing, other subtle ways of enzyme inactivation during the course of evolution can explain those results (50). At this time, we cannot discard the alternative hypothesis that A. thaliana L subunits from groups I and II are catalytic but could not interact properly with APS2. In the literature, no other L subunit from these two groups has been expressed and tested for intrinsic activity. Based on sequence, we cannot rule out the possibility of finding a catalytically competent subunit among groups I and II.
Phylogenetic trees based on protein sequences are very powerful tools to test evolutionary hypothesis, but they do not show the functional information that those sequences carry. Gene synthesis of hypothetical ancestors and characterization of their products may tackle this problem (51). As performed in this paper, analyses of structure-function relationships should complement phylogenetic analysis for tracing and understanding how and when certain former enzymes acquired new roles and lost catalytic function. The use of this information could potentially increase the accuracy of function prediction for these difficult cases (50,52).
Enzyme Inactivation as a General Strategy-Divergence from an ancestral catalytic subunit to form hetero-oligomers with non-catalytic and catalytic subunits could have been a strategy used by nature to add complexity and explore new regulatory functions. There are phylogenetic evidences suggesting this could be a general evolutionary mechanism. In a superfamily of enzymes, if only one branch ("odd one out") in the tree is non-catalytic, the common ancestor most probably was catalytic as the majority of the branches. Six cases of hetero-oligomers with solved three-dimensional structures that meet this criterion have been observed (50). In this work, we experimentally tested this hypothesis for a plant ADP-Glc PPase and found that the common ancestor of catalytic and non-catalytic subunits was catalytic. This enzyme is a very attractive model because the "inactivated" subunit acquired differential modulatory properties in plant tissues, expanding the plasticity for regulation of a key synthetic pathway. After gene duplication, the ADP-Glc PPase subunits acquired asymmetric functions. The S subunit remained catalytic with defective allosterism whereas the L subunit became modulatory (i.e. modifying the properties of the S subunit) and lost catalytic function.
Considering the ancient inactivation of L subunits and their higher divergence (Fig. 3), it is remarkable that the general architecture of their substrate site has not changed much by evolution. The wild-type enzyme and S D145N L K44R/T54K had comparable apparent affinities for the substrates. This preservation occurred despite substrate binding does not seem to be an adaptive pressure because is not needed for the modulatory roles of the L subunit. It was shown previously that mutations in the L subunit of a critical residue for Glc1P binding (Lys 213 ) do not affect the modulatory properties of the potato tuber ADP-Glc PPase (9). In addition, in the wheat endosperm enzyme, Gln has replaced that Lys. Perhaps, the active-site architecture has been kept as a by-product of evolution or spandrel (53) because some disruptions could indirectly affect the modulatory properties. For instance, Asp 160 in the wild-type L subunit is conserved in all ADP-Glc and other nucleotide-sugar PPases (6,25,44). In the catalytic subunits, it has a catalytic role but replacement in the wild-type L subunit of Asp 160 altered the modulatory properties of the tetramer (6). This residue, Asp 160 , "recovered" its critical role for activity in the L K44R/T54K mutant (Table IV).
Adaptation from enzymes to non-enzymes and vice versa is a very exciting topic in protein evolution because it involves the generation of novel functions (50). Not only subunits from hetero-oligomers, but also many enzymes and non-enzymes homologues, are being detected. Based on the odd one out criterion, it has been reported some cases of non-enzymes that became enzymes, but the opposite scenario was observed more FIG. 3. Phylogenetic relationship of ADP-Glc PPases from photosynthetic eukaryotes. The unrooted tree was built as described under "Experimental Procedures" and reference numbers are listed on Supplemental Table I. All the ADP-Glc PPases are from angiosperms because there are no complete sequences available from gymnosperms. Dotted lines separate L from S subunits, whereas the cyanobacterial ADP-Glc PPase from Anabaena was included as outgroup. The black arrow points to the potato tuber L subunit analyzed in this study (sequence 66). The general topology of the tree had a strong support by a maximum-likelihood method (Supplemental Fig. 1). Between brackets, it is indicated the residues conserved in each group in the positions that correspond to Lys 44 (first bracket) and Thr 54 (second bracket) in the potato tuber L subunit. The scale on the bottom left represents the number of substitutions per site.
often (50,54). Some examples may be critical for the appearance of higher levels of complexity and control. For instance, a thrilling proposal is that many transcription regulators have an ancient catalytic past (55,56). Supporting experimental evidence, as performed in the present study, will be enlightening to confirm these hypotheses and understand how key regulatory roles have evolved at a molecular level.