Introduction

Top Abstract Introduction Materials and Methods Results Discussion References

High-throughput genome sequencing has dramatically expanded our knowledge of the proteins of the natural world. Next steps in understanding these macromolecules will involve studies of biochemical and biological function, depending critically on three-dimensional structural information. Nascent structural genomics efforts are aimed at developing experimental and computational pipelines for studying protein structure and integrating their results into the mainstream of biomedical research (1).

The New York Structural Genomics Research Consortium (http://www.nysgrc.org/, including Albert Einstein College of Medicine, Brookhaven National Laboratory, Mount Sinai School of Medicine, The Rockefeller University, and Weill Medical College of Cornell University) has been conducting a structural genomics pilot study under the auspices of the National Institutes of Health Protein Structure Initiative (http://www.nih.gov/nigms/funding/psi.html). Initial effort focused on proteins from Saccharomyces cerevisiae, an intensively studied eukaryotic organism with a fully sequenced genome that contains numerous human gene homologues. Target selection was aided by the results of automated comparative protein structure modeling by using the S. cerevisiae genome (2). We focused on proteins for which no structural information was available, with emphasis on members of large protein families that would permit homology modeling of as many related proteins as possible. Among the first 111 targets selected for x-ray crystallographic structure determination were two enzymes [Target 100, mevalonate-5-diphosphate decarboxylase (MDD), and Target 109, isopentenyl disphosphate isomerase (IDI)] from the sterol/isoprenoid biosynthesis pathway (ref. 3; Fig. 1).

View larger version (41K): [in this window] [in a new window]   Fig. 1.   Pathways for biosynthesis of isopentenyl diphosphate. Isopentenyl diphosphate, the central intermediate in sterol/isoprenoid biosynthesis, is produced by two independent pathways, which have different evolutionary distributions (39, 40).

	Materials and Methods

Top Abstract Introduction Materials and Methods Results Discussion References

Protein Expression/Purification. DNA fragments encoding MDD (S. cerevisiae gene mvd1, ORF YNR043W, 396 residues) and IDI (S. cerevisiae gene idi1, ORF YPL117C, 288 residues) were amplified from the genome of S. cerevisiae strain S288C by PCR and cloned into a T7 RNA polymerase-dependent Escherichia coli expression vector (4) that furnished a protease-cleavable N-terminal His₆ tag. Gel electrophoresis of cell extracts demonstrated that expressed MDD and IDI were both in the soluble fraction. Nickel-ion affinity chromatography and gel filtration yielded highly purified proteins, as judged by mass spectrometry. After difficulties with S. cerevisiae IDI crystallizability, E. coli IDI (gene idi, 182 residues) was cloned from UT5600 genomic DNA, expressed as a glutathione S-transferase fusion, and purified to homogeneity by using published procedures (5).

Crystallizability Screening and Crystallization. Conformationally stable, monodisperse macromolecular preparations represent good crystallization candidates (6, 7). Conformational heterogeneity can be detected by using fluorescence and circular dichroism spectroscopy or proteolysis combined with MS (8). Polydispersity (nonspecific aggregation) can be detected by using dynamic light scattering (9). S. cerevisiae MDD behaved as a monodisperse dimer with an apparent molecular mass of 110 kDa (predicted mass for the dimer is 94 kDa) in aqueous solution. S. cerevisiae IDI was aggregated under similar conditions and was abandoned in favor of the more suitable homologue from E. coli, which behaves as a monodisperse monomer.

Data Collection and Structure Determination. X-ray diffraction data for MDD and IDI were recorded under standard cryogenic conditions at Beamline X25 at the National Synchrotron Light Source and Beamline F2 at the Cornell High Energy Synchrotron Source, respectively, and processed by using DENZO/SCALEPACK (10). Structures of MDD and IDI were obtained with the multiwavelength anomalous dispersion method (11) from diffraction data recorded at 3 (MDD) or 2 (IDI) x-ray wavelengths (Table 1). Selenium atomic positions (9 and 6 selenium sites, respectively) were located with SNB (12). MLPHARE phasing followed by density modification (13) yielded high-quality experimental phases for both systems. Electron density map interpretation (14) and refinement (15) produced structures of MDD (391 residues, 164 waters, R factor = 23.9%, R free = 26.8%, 2.3 Å resolution) and IDI (354 residues, 261 waters, R factor = 22.9%, R free = 27.9% 2.5 Å resolution).

Multiple Sequence Alignments. Proteins similar to S. cerevisiae MDD and E. coli IDI (E value <10⁴) were identified by using -BLAST (16). Multiple sequence alignments were prepared by using CLUSTAL (17), and conservation was calculated with BLOSUM62 (18). Alignments of selected MDDs and IDIs are given in Figs. 2 and 3, respectively.

View larger version (76K): [in this window] [in a new window]   Fig. 2.   MDD sequence alignment. Secondary structural elements of S. cerevisiae MDD are shown with cylinders (-helices) and arrows (-strands). Gray dots denote poorly resolved residues in the final electron density map. Color-coding denotes sequence conservation among MDDs (white  green ramp, 30  100% similarly). Red box denotes the putative ATP-binding P loop.

View larger version (40K): [in this window] [in a new window]   Fig. 3.   IDI sequence alignment. Secondary structural elements, poorly resolved residues, and sequence conservation are denoted as in Fig. 2. #, metal-binding residues; *, conserved residues in the cleft; o = the active-site Cys. Some of the N- and C-terminal residues of IDI sequences other than E. coli have been excluded for clarity.

Homology Modeling. Automated comparative protein structure modeling (hereafter referred to as "homology modeling") with MODPIPE (19) was carried out by using each experimentally determined structure (hereafter referred to as "structures") as a modeling template. Candidates for modeling were obtained from the National Center for Biotechnology Information (ftp://ftp.ncbi.nlm.nih.gov/blast/db) by using an iterative -BLAST search (E value <100 for at least 30 residues) with the sequence of the modeling template. Homology models were computed with the -BLAST alignments by means of satisfaction of spatial restraints, as implemented in MODELLER (20), and assessed by computing a model score that uses a statistical energy function, sequence similarity with the modeling template, and a measure of structural compactness (2). Tests with known structures have shown that models with scores from 0.7 to 1.0 have the correct fold at a 95% confidence level (2). All models discussed below have scores >0.7.

	Results

Top Abstract Introduction Materials and Methods Results Discussion References

MDD Structural Overview. MDD is a single / domain (Fig. 4A, dimensions 75 Å × 42 Å × 38 Å) with a deep cleft. The order of secondary structural elements is 1-2-3-4-5-6-A-7-B-C-D-8-9-10-E-F-G-H-11-12-I-J-13-14, and the structure consists of three antiparallel -sheets (1-6-7-14, 2-5-8-9, 10-11-12-13) and three sets of helices (A-D, E-G, H-J). The crystallographic two-fold symmetry axis parallel to the c axis generates a symmetric dimer, which presumably corresponds to the oligomerization state detected by dynamic light scattering.

View larger version (33K): [in this window] [in a new window]   Fig. 4.   S. cerevisiae MDD and M. jannaschii HSK. Ribbon drawings of MDD (A) and HSK (B) in the same orientation. The two-fold rotational symmetry axis that generates the MDD homodimer is indicated. Putative P loops are colored red.

Homology Modeling with MDD. Comparison of S. cerevisiae MDD with the Protein Data Bank (23) in November 2000 revealed no similar structures, as judged by DALI (24). Homology modeling with S. cerevisiae MDD gave models for the other MDDs plus various GHMP kinases (21), including the canonical galactokinases (GK), homoserine kinases (HSK), mevalonate kinases (MK; Fig. 1), and phosphomevalonate kinases (PMK; Fig. 1), plus diphosphocytidyl-2-C-methyl-D-erythritol kinases [CMK (EC 2.7.1.-), an enzyme in the 1-deoxy-D-xylulose-5-phosphate pathway to isopentenyl diphosphate; Fig. 1], and some hypothetical proteins. A kinase mechanism of action for MDD is not unreasonable. MDD phosphorylates the substrate C-3 hydroxyl group, followed by elimination of both the added phosphate and carboxylate groups to give isopentenyl diphosphate, ADP, and P_i. Our modeling results suggest that all three enzymes responsible for sequential conversion of mevalonate to isopentenyl diphosphate (MK, PMK, and MDD; Fig. 1) have the same fold, indicating that they arose from a common precursor and may represent an example of retrograde evolution (25). It is remarkable that diphosphocytidyl-2-C-methyl-D-erythritol kinase (CMK), which is part of the mevalonate-independent pathway for sterol/isoprenoid biosynthesis in plastids and bacteria, also seems to have evolved from the same ancestral small-molecule kinase.

IDI Structural Overview. IDI is a single / domain (Fig. 5A, dimensions 28 Å × 40 Å × 27 Å) with a divalent metal ion bound within a deep cleft. The order of secondary structural elements is 1-2-A-3-4-5-6-7-B-8-9-10-11-12-C-D-E-F, and the structure consists of two mixed polarity (2-1-3-10-9, 7-4-11-8) and one anti-parallel (12-5-6) -sheets. The second and third -sheets are backed by -helices.

View larger version (28K): [in this window] [in a new window]   Fig. 5.   E. coli IDI and MutT. Ribbon drawings IDI (A) and MutT (B) in the same orientation. Ball-and-stick representations of the divalent metal ion and putative active-site residues are given for IDI (atom type code: C, green; S, yellow; N, blue; O, red; Mn2+, gray). The positions of the equivalent conserved motifs of the two proteins are colored magenta.

IDI Resembles MutT. Comparison of our IDI structure with the contents of the Protein Data Bank (PDB) in January 2001 revealed one relative, E. coli MutT (PDB ID code 1TUM; 108 -carbon pairs, rmsd = 3.0 Å, 13% identity, Z score = 6.7). The solution NMR structure of MutT (Fig. 5B) resembles the portion of IDI containing the 3-10-9 and 7-4-11-8 sheets and two of the flanking -helices (B and F).

Homology Modeling with IDI and MutT. By using the structures of IDI and MutT as modeling templates, MODPIPE produced models (as of April 2001) for 202 nonredundant protein sequences (length > 100 residues)62 with both templates, 103 with IDI only, and 37 with MutT only. The MutT model derived from IDI vs. the solution NMR structure of MutT gave an rmsd of 3.0 Å (69 -carbon pairs, 18% identity, Z score = 2.7). No model for IDI (score > 0.7) was produced with the structure of MutT as a template. In cases where models for the same sequence were obtained from both structural templates (IDI and MutT), the average rmsd between alternative models was 3.6 Å with <Z score> = 5.6 ± 1.0. This <Z score> is considerably lower than the corresponding value obtained by comparing MDD- and HSK-derived alternative models, and presumably reflects the greater structural difference between IDI and MutT. Nonetheless, our structural comparison and modeling results do confirm the recently proposed similarity of the IDIs and the nudix proteins (32).

	Discussion

Top Abstract Introduction Materials and Methods Results Discussion References

One of the goals of the New York Structural Genomics Research Consortium is to generate experimentally determined structures with sufficient diversity to support homology modeling of large numbers of protein sequences. Recent analyses by Vitkup et al. (33) have suggested that determination of as few as 16,000 selected structures could enable production of "accurate" homology models for 90% of all proteins found in nature (see below).

Homology modeling can be distinguished from all other methods for analyzing relationships among protein sequences because it yields atomic coordinates suitable for direct comparisons with x-ray and solution NMR structures. Acceptable models can be divided into three accuracy classes, characterized with blind tests by using known structures (34). Models based on >50% sequence identity with the template are comparable in accuracy to 3 Å resolution x-ray structures or medium-resolution solution NMR structures (10 long-range restraints per residue). Models obtained with less similar templates (30-50% identity) typically have >85% of their -carbons within 3.5 Å of the correct position. When sequence identities fall below 30%, models with acceptable model scores (>0.7) may contain significant errors arising from ambiguities in the sequence alignment of the modeling candidate with the template, but can nevertheless be used for protein fold identification.

Targets 100 (MDD) and 109 (IDI) were selected for experimental structure determination to provide modeling templates for two enzyme families in the sterol/isoprenoid biosynthesis pathway and, indeed, we obtained models for 21 MDDs and 36 IDIs. Greater modeling coverage was, however, possible because both MDD and IDI proved to be members of distinct protein superfamilies encompassing a range of different enzyme activities and proteins of unknown biochemical function. Even wider coverage was afforded by the availability of a second modeling template from each of these protein sequence/structure families. Of the 379 sequences modeled with one or more of the MDD, HSK, IDI, and MutT templates, 3 (<1%) fall into the highest accuracy range (>50% identity), 53 (14%) in the medium accuracy range (30-50% identity), and 323 (85%) in the lowest accuracy range (<30% identity). Virtually all of the 56 proteins with medium or highest accuracy models (30% identity) are thought to have the same enzymatic function as the template (with two possible exceptions in the nudix/IDI superfamily). In clustering the members of these two superfamilies into groups of the most closely related proteins (by means of sequence comparisons), proteins of unknown function were often grouped with proteins of known enzymatic activity, allowing tentative assignment of biochemical function (Tables 2-5). Isolated cases of patently incorrect functional annotation were also found (Tables 2-5).

The modeling coverage of the GHMP kinase superfamily provided some insights into the problem of target selection for structural genomics. The structure of S. cerevisiae MDD produced medium or high accuracy models for 80% of the modeled sequences thought to have MDD activity, but coverage of the HSK family provided by the M. jannaschii HSK template was not as broad (only 41% of HSK models are medium or high accuracy). Current GHMP kinase superfamily members can be clustered into 19 discrete subfamilies with at least 4 members, encompassing the MDDs, 3 clusters of GKs, 4 clusters of HSKs, 3 clusters of MKs, the PMKs, the D-glycero-D-manno-heptose 7-phosphate kinases, 3 clusters of diphosphocytidyl-2-C-methyl-D-erythritol kinases (CMKs), the archael shikimate kinases, and 2 distinct clusters of hypothetical proteins. We estimate that 17 additional experimentally determined structures will be required to produce medium to high accuracy models for most members (80%) of each detectable GHMP kinase subfamily. The nudix/IDI superfamily, on the other hand, seems to have much greater structural diversity, and achieving equivalent coverage by medium to high accuracy models may present a greater logistical challenge.

Structures of GHMP kinase superfamily members and resulting homology models may guide experimentation aimed at defining enzymatic function, cofactor requirements, and mechanism(s) of action, which should allow us to understand better how this family of structurally similar yet functionally diverse enzymes evolved from a common ancestor. The GHMP kinase models may be of some medical relevance in understanding the structural bases of a number of diseases caused by single-nucleotide polymorphisms in coding regions. Impairment of human GK function by missense mutations within the enzyme leads to galactosemia and cataract formation in newborns (35), which can be reversed by restricting dietary galactose. Mutations in the human gene encoding MK lead to either hyperIgD and periodic fever syndrome or mevalonic aciduria, a severe and frequently fatal malady (36, 37). Finally, mutations in both MK and PMK have been implicated in the development of Zellweger syndrome, a genetic disorder characterized by defects in peroxisome biogenesis (38).