Distinct Conformation of ATP Molecule in Solution and on Protein.

Adenosine triphosphate (ATP) is a versatile molecule used mainly for energy and a phosphate source. The hydrolysis of γ phosphate initiates the reactions and these reactions almost always start when ATP binds to protein. Therefore, there should be a mechanism to prevent spontaneous hydrolysis reaction and a mechanism to lead ATP to a pure energy source or to a phosphate source. To address these questions, we extensively analyzed the effect of protein to ATP conformation based on the sampling of the ATP solution conformations obtained from molecular dynamics simulation and the sampling of ATP structures bound to protein found in a protein structure database. The comparison revealed mainly the following three points; 1) The ribose ring in ATP molecule, which puckers in many ways in solution, tends to assume either C2' exo or C2' endo when it binds to protein. 2) The adenine ring in ATP molecule, which takes open-book motion with the two ring structures, has two distinct structures when ATP binds to protein. 3) The glycosyl-bond and the bond between phosphate and the ribose have unique torsion angles, when ATP binds to protein. The combination of torsion angles found in protein-bound forms is under-represented in ATP molecule in water. These findings suggest that ATP-binding protein exerts forces on ATP molecule to assume a conformation that is rarely found in solution, and that this conformation change should be a trigger for the reactions on ATP molecule.

Adenosine triphosphate (ATP) is a widely used molecule in the cell for an energy source1. A textbook example of the use of ATP is a chemical bond formation between two substrates coupled with ATP hydrolysis catalyzed by an enzyme. In this reaction, a phosphoanhydride bond between β and γ phosphate groups is cleaved, and the released energy is used to condense the substrates. The released energy can also be a trigger for alteration of the conformation of protein2. In either case, the remaining adenosine diphosphate (ADP) and the inorganic phosphate are released to water. Some of the reactions yield an inorganic diphosphate by cleaving the bond between α and β phosphate groups3. Other than the reaction to gain energy, ATP is utilized as a source for phosphate group, adenosine monophosphate (AMP) and adenine. These chemical groups are utilized for phosphorylation that transfers the inorganic phosphate to the substrate4, adenylation that transfers AMP to the substrate5, and adenosylation that transfers adenosyl to the substrate6, respectively.

The use of the same ATP molecules in a variety of chemical reactions is evidently based on its versatility in the conformation, but the mechanism for regulating the conformation for distinct functions has not been addressed. The ATP molecule that undertakes a hydrolysis between β and γ phosphate groups, for instance, should block the chemical reaction pathways to phosphorylation, adenylation and others, otherwise the unrelated functions would be carried out. In addition, ATP molecule in water needs to have a certain mechanism to stay away from the chemically reactive situations leading to a spontaneous hydrolysis.

These conjectures can be tested by protein structure database analysis and computer simulation. Accumulation of the coordinate data of ATP bound to the proteins enabled us to obtain ATP conformations on proteins at the variety of functions. Improvements in simulation techniques and computer hardware enable us to sample conformations of ATP in water. Comparisons of these ATP conformations will give us a clue to solidify the conjecture.

Here, we compared the structures of ATP molecules in Protein Data Bank (PDB)7 and those sampled from the molecular dynamics (MD) simulation. We found that the conformation of protein-bound ATP is under-represented in ATP in water, which suggests that ATP molecule should be forced to take a specific conformation on a protein to initiate biological functions.

Methods

Choosing proteins with ATP molecule from PDB

Three-dimensional coordinate data of protein structure with ATP were selected from PDB7. The protein entries with coordinates of ATP were first selected on Het-PDB Navi.8 using “ATP” as a query term. Redundancy in entries was eliminated by grouping the proteins with their sequence identity. The interactions between protein chain and ATP molecule were detected by differences in accessible surface areas of the protein chain when the area was calculated with and without the ATP molecule. We calculated the accessible surface area by the in-house program and the program is now available at http://cib.cf.ocha.ac.jp/bitool/ASA/. The calculation is based on the method of Shrake and Rupley9. Chains with less than 60 amino acid residues were discarded. Classification of proteins by sequence identity was carried out using BLASTClust10. The sequence identity for the classification was set to 25%. From each group, a protein chain with the best resolution was selected as the representative.

Conformation sampling of ATP molecule by molecular dynamics simulation

Molecular dynamics simulation of ATP was performed to sample conformations of ATP in water. The initial structure of ATP was taken from the three-dimensional structure data of Thermus thermophilus D-alanine:D-alanine ligase (PDB ID, 2zdq)11. The ATP numbered 1501 in A chain was used. For the calculation, GROMACS 4.0.412 was used. We employed a standard NPT procedure for the simulation described in the manual of GROMACS. We used the force field for ATP molecule implemented in ffG43a1.rtp file. The file described the parameters for all the atoms of ATP except for methyl hydrogen atoms, which were united to the bonded carbon atoms. A hydrogen atom not described in PDB file was geometrically generated at an allowed position. The geometric center of the ATP molecule was then placed at the center of a cube with 2.7×104 Å3 volume filled with water molecules with periodic boundary condition. By removing water molecule overlapping with ATP, the number of water molecules was settled to 876. After minimizing the energy of the system by steepest decent method and performing molecular dynamics with restraint on ATP in 1 ns, we performed 2 ns simulation of ATP in solvent with 2 fs step size. The temperature was set in 300 K. Cutoff distance of van der Waals and electrostatic interactions was set to 10 Å. We ran ten different sets of the simulation starting with a different random-number seed. From each trajectory file, coordinates of ATP in every 0.1 ps were retrieved and snapshot structures from the latter 1 ns simulation were used for analyses.

Comparison of ATP structures: torsion angle and ring curvature

Conformations of ATP molecules in protein-bound and free forms were compared by torsion angles of bonds and flatness of ring structures.

Torsion angles in ATP were defined as shown in Figure 1. The definition is the same as the ones commonly used in DNA and RNA (see Chapter 5 of Schlick T.13, for instance). A torsion angle of a glycosyl bond (C1′-N9), for example, is defined by O4′, C1′, N9 and C4. The cis position of O4′ and C4 is defined as zero degree and the clockwise rotation of the N9-C4 bond viewed in C1′-N9 direction is defined as a positive rotation.

Figure 1

Definition of the torsion angle for ATP molecule. Each torsion angle is named as shown in the right box. In the box, the torsion angle of the bond by the second and the third atoms is defined by the rotation between the first-second and the third-fourth bonds. cis location of the first and the fourth atoms is defined as zero degree. The order of the atoms also defines the sign of rotation, namely clockwise rotation of the fourth atom against the first atom is defined as a positive rotation. The arrows in the figure depict the positive rotation of the bond.

Flatness of the ring structure of ribose and adenine was calculated using discrete Gaussian curvature (K) and mean curvature (H) descriptions (Fig. 2). The Gaussian curvature at a point on a surface is defined as a product of the maximum and minimum curvatures of a plane embedding the normal vector of the point (the principal curvatures), and the mean curvature is defined as a mean of the principal curvatures. With both curvatures, the degree of flatness and of puckering of a ring structure can be described. Here we employed the definition of the discrete Gaussian and mean curvatures described in references 14 and 15. The discrete Gaussian curvature at the gravity center of a ring can be calculated as; and the discrete mean curvature at the gravity center can be calculated as; In these calculations, A is the area of triangle spanned by atoms i, j and the gravity center of the ring, θ is the angle in radian between the two lines, the line connecting atom i and the gravity center, and the line connecting atom j and the gravity center, δ is the torsion angle in radian between two triangles over the line drawn between atom i and the gravity center, and l is the length of the line drawn between atom i and the gravity center. The subscripts i and j go over all the atoms for the ring14,15. For the curvature calculation of the ribose, namely C1′, C2′, C3′, C4′ and O4′ atoms were used. Curvature calculation for the two ring structures in adenine was done separately. For the curvature calculation of the six-membered ring in the adenine, N1, C2, N3, C4, C5 and C6 atoms were used, and of the five-membered ring, C4, C5, N7, C8, and N9 atoms were used. The relative orientation of the two rings in the adenine was described by the flatness of the pseudo-hexagon consisting of C6, C5, N7, N9, C4 and N3 atoms.

Figure 2

The definition of the discrete Gaussian and mean curvatures at the gravity center.

Intuitively, the discrete Gaussian curvature measures whether the surface is curved or not, whilst the discrete mean curvature measures the degree of the mixture of the concaveness and convexness. In this analysis, the Gaussian curvature at the gravity center of the ring is always negative. The sign of the mean curvature depends on the strength of concaveness and convexness of the ring structure at the gravity center. Flatness and puckering of the ring can be described by both curvatures through concaveness and convexness.

Results and Discussion

Coordinate set of ATP from PDB

The set of proteins with ATP in PDB is shown in Table 1. There were 188 unique protein-ATP complex. The uniqueness was defined by the sequence identity of the proteins. No proteins in the set have sequence identity more than 25% based on the calculation by BLASTClust10. The biological uniqueness of these proteins was checked based on Uni Prot16 ID. UniProt ID is basically built by protein function abbreviation with a species name abbreviation connected by an underscore. None of the entries in Table 1 has the same protein function based on the UniProt ID.

Table 1

Functional classification of ATP-binding proteins

ATP hydrolysis, Pi is released (energy extration reaction)
Protein Name	Family	PDB ID	chain	resol	Uniprot ID
vacuolar protein sorting-associating protein 4B	AAA ATPase family	2zan	A	3.00	VPS4B_MOUSE
N-ethylmaleide sensitive factor	AAA ATPase family	1nsf	A	1.90	NSF_CRIGR
FbpC nucleotide-binding domain	ABC transporter domain	3fvq	B	1.90	FBPC_NEIG1
histidine permease	ABC transporter superfamily	1b0u	A	1.50	HISP_SALTY
maltose/maltodextrin transport ATP-binding protein MalK	ABC transporter superfamily	1q12	A	2.60	MALK_ECOLI
ATP-binding cassette sub-family B meber 6	ABCB family	3nh9	A	2.10	ABCB6_HUMAN
alpha actin 1	actin family	2fxu	A	1.35	ACTS_RABIT
actin-related protein 2	actin family	1tyq	B	2.55	ARP3_BOVIN
arsenical Pump-driving ATPase	arsA ATPase family	1ii0	B	2.40	ARSA1_ECOLI
ATP synthase subunit alpha	ATPase alpha/beta chains family	2r9v	A	2.10	ATPA_THEMA
v-type ATP synthase beta chain	ATPase alpha/beta chains family	3b2q	A	2.10	VATB_METMA
biotin carboxylase	biotin carboxylation domain	1dv2	A	2.50	ACCC_ECOLI
sarcoplasmic/endoplasmic reticulumn calcium ATPase 1	cation transport ATPase (P-type) family	3ar4	A	2.15	AT2A1_RABIT
GroEL	chaperonin (HSP60) family	1kp8	A	2.00	CH60_ECOLI
heat shock locus U (HslU)	clpX chaperone family	1do0	A	3.00	HSLU_ECOLI
DNA mismatch repair protein Mlh1	DNA mismatch repair mutL/hexB family	3na3	A	2.50	MLH1_HUMAN
DNA mismatch repair protein MutS	DNA mismatch repair mutS family	1w7a	A	2.27	MUTS_ECOLI
PurL, Formylglycinamide ribonucleotide amidotransferase	FGAMS family	2hs0	A	2.52	PURL_THEMA
Gar synthetase (PurD)	GARS family	2yw2	A	1.80	PUR2_AQUAE
aspartyl/glutamyl-tRNA amidotransferase subunit B	gatB/gatE family	3h0r	H	3.00	GATB_AQUAE
70kDa heat shock cognate protein	heat shock protein 70 family	1kax	A	1.70	HSP7C_BOVIN
PcrA DNA helicase	helicase family	1qhh	B	2.50	PCRA_BACST
nitrogenase iron protein 1	nifH/bchL/chlL family	2c8v	A	2.50	NIH1_AZOVI
cell division inhibitor MinD	parA family	3q9l	A	2.34	MIND_ECOLI
bacterial chromosome segregation protein SoJ	ParAB family	2bek	A	1.80	Q72H90_THET2
5-formaminoimidazole-4-carboxamide-1-beta-D-ribofuranosyl 5′-monophosphate synthetase	phosphohexose mutase family	2r7l	A	2.10	PURP_METJA
phosphoribosylaminoimidazole carboxylase ATPase subunit	purK/purT family	3eth	A	1.60	PURK_ECOLI
glycinamide ribonucleotide transformylase (purT)	purK/purT family	1kj9	B	1.60	PURT_ECOLI
Holliday junction DNA helicase RuvB	ruvB family	1j7k	A	1.80	RUVB_THEMA
phoshpribosylamidoimidazole-succinocarboxamide synthase	SAICAR synthetase family	1obd	A	1.40	PUR7_YEAST
translocase SecA subunit	secA family	2fsg	B	2.20	SECA_ECOLI
larget T antigen helicase domain	SF3 helicase domain	1svm	C	1.94	LT_SV40
Psp operon transcriptional activator (PspF)	sigma-54 factor interaction domain	2c96	A	1.80	PSPF_ECOLI
Rad50 ABC-ATPase N-terminal domain	SMC family	1f2u	A	1.60	RAD50_PYRFU
sulfiredoxin	sulfiredoxin family	3cyi	A	1.80	SRXN1_HUMAN
NTPase P4 (molecular motor)	superfamily 4 helicase motif	2vhq	A	2.15	Q94M05_9VIRU
transglutaminase 2	Transglutaminase family	3ly6	A	3.14	TGM2_HUMAN
EcoR124I restriction enzyme HSDR subunit	typeII restriction enzyme	2w00	B	2.60	Q304R3_ECOLX
UvrABC component UvrB	uvrB family	1d9z	A	3.15	UVRB_BACCA
twitching motility protein PilT	not classified	2eww	A	3.20	O66950_AQUAE
transcriptional regulatory protein ZraR	not classified	1ojl	E	3.00	ZRAR_SALTY
myosin II heavy chain	not classified	1fmw	A	2.15	MYS2_DICDI
dethiobiotin synthetase	not classified	1a82	A	1.80	BIOD_ECOLI
ATP hydrolysis, Pi is transferred (phosphprylation)
Protein Name	Family	PDB ID	chain	resol	Uniprot ID
isocitrate dehydrogenase kinase/phosphatase (AceK)	AceK family	3eps	A	2.80	ACEK_ECO57
cAMP-dependent protein kinase	AGC Ser/Thr protein kinase family	3fjq	E	1.60	KAPCA_MOUSE
protein kinase C iota type	AGC Ser/Thr protein kinase family: PKC subfamily	3a8w	B	2.10	KPCI_HUMAN
G protein coupled receptor kinase 1 (crystals of 6 different states)	AGC Ser/Thr protein kinsae family: GRK kinase family	3c4w	B	2.70	RK_BOVIN
myosin heavy chain kinase A	alpha-type protein kinase family	3lmi	B	2.20	MHCKA_DICDI
Isopentenyl phosphate kinase	Amino acid kinase family	3ll5	C	1.99	Q9HLX1_THEAC
anti-sigma F factor	anti-sigma-factor family	1tid	A	2.50	SP2AB_BACST
ribokinase	carbohydrate kinase pfkB family	3ikh	A	1.88	A6T989_KLEP7
casein kinase-1	CK1 Ser/Thr protein kinase family	1csn	A	2.00	CKI1_SCHPO
dephospho-CoA kinase	coaE family	1jjv	A	2.00	COAE_HAEIN
mevalonate kinase	GHMP kinase family	1kvk	A	2.40	KIME_RAT
gluconate kinase	gluconokinase gntK/gntV family	1ko5	A	2.28	GNTK_ECOLI
Inositol 1,4,5-triphosphate 3-kinase B	inositol phosphokinase (IPK) family	2aqx	A	2.50	IP3KB_RAT
KaiC	kaiC family	2gbl	A	2.80	KAIC_SYNP7
l-seryl-tRNA kinase	L-seryl-tRNA(Sec) kinase family	3am1	A	2.40	PSTK_METJA
NAD kinase	NAD kinase family	1z0s	A	1.70	PPNK_ARCFU
nucleotide diphosphate kinase	NDK family	1wkl	B	2.20	NDK_THET8
pyruvate dehydrogenase kinase isoform 2	PDK/BCKDK protein kinase family	2bu2	A	2.40	PDK2_HUMAN
phosphoenolpyruvate carboxykinase	phosphoenolpyruvate carboxykinase family	2olr	A	1.60	PPCK_ECOLI
phosphofruktokinase	phosphofructokinase family	3o8l	A	3.20	K6PF_RABIT
phosphoglycerate kinase	phosphoglycerate kinase family	1vjd	A	1.90	PGK1_PIG
phosphatidylinositol 3-kinase catalytic subunit	PI3/PI4-kinase family	1e8x	A	2.20	PK3CG_PIG
polyhosphate kinase	polyphosphate kinase family	1xdp	A	2.50	PPK_ECOLI
Pantothenate kinase	prokaryotic pantothenate kinase family	2zsf	A	2.80	COAA_MYCTU
cell division protein kinse 2	protein kinase superfamily	2cch	A	1.70	CDK2_HUMAN
pyridoxine kinase	pyridoxine kinase family	2ddo	A	2.60	PDXK_ECOLI
pyruvate kinase	pyruvate kinase family	1a49	A	2.10	KPYM_RABIT
Rio1 serine kinase	RIO-type Ser/Thr kinase family	1zp9	A	2.00	RIO1_ARCFU
Rio2 serine kinase	RIO-type Ser/Thr kinase family	1zao	A	1.84	RIO2_ARCFU
mitotic checkpoint serine/threonin-protein kinase Bub1	Ser/Thr protein kinase family	3e7e	A	2.31	BUB1_HUMAN
SR protein kinase	Ser/Thr protein kinase family	1q97	A	2.30	SKY1_YEAST
shikimate kinase	shikimate kinase family	2iyw	A	1.85	AROK_MYCTU
Tao2 kinase domain	STE20 subfamily	1u5r	A	2.10	TAOK2_RAT
thymidylate kinase	thymidylate kinase family	1e2q	A	1.70	KTHY_HUMAN
thiazole kinase	Thz kinase family	1esq	C	2.50	THIM_BACSU
MET receptor tyrosine kinase	Tyr protein kinase family	3dkc	A	1.52	A1L467_HUMAN
phosphofruktokinase	not classified	3f5m	B	2.70	O15648_9TRYP
D-alanine-D-alanine ligase	not classified	2zdq	A	2.30	Q5SHZ3_THET8
chloramphenicol phosphotransferase	not classified	1qhx	A	2.50	CPT_STRVL
aminoglycoside phosphotransferase	not classified	3hav	B	2.45	Q9EVD7_ENTFC
Thiamine monophosphate kinase	not classified	3c9r	A	2.30	O67883_AQUAE
UMP kinase	not classified	2jjx	A	2.82	Q81S73_BACAN
ATP hydrolysis, PPi is released (energy extration reaction)
Protein Name	Family	PDB ID	chain	resol	Uniprot ID
adenylate cyclase type 5	adenylyl cyclase class-4/guanylyl cyclase family	3c16	A	2.87	ADCY5_CANFA
argininosuccinate synthetase	argininosuccinate synthase family	1kp3	A	2.00	ASSY_ECOLI
beta-lactam synthetase	asparagine synthetase family	1mb9	B	2.11	BLS_STRCL
Acyl-coenzyme A synthetase Acsm2A	ATP-dependent AMP-binding enzyme	3c5e	A	1.60	ACS2A_HUMAN
D-alanine-polyphosphoribitol ligase subunit 1	ATP-dependent AMP-binding enzyme family	3fce	A	1.90	DLTA_BACCR
DNA ligase from bacteriophage T7	ATP-dependent DNA ligase family	1a0i	A	2.60	DNLI_BPT7
tryptophan-tRNA synthetase	class-I aminoacyl-tRNA synthetase familiy	1mau	A	2.15	SYW_BACST
glutamyl-tRNA synthetase	class-I aminoacyl-tRNA synthetase familiy	1j09	A	1.80	SYE_THET8
glutaminyl-tRNA synthetase	class-I aminoacyl-tRNA synthetase family	1gtr	A	2.50	SYQ_ECOLI
tyrosine-tRNA synthetase	class-I aminoacyl-tRNA synthetase family	1h3e	A	2.90	SYY_THETH
tryptophanyl-tRNA synthetase	class-I aminoacyl-tRNA synthetase family	2qui	A	2.40	SYWC_HUMAN
histidyl-tRNA synthetase	class-II aminoacyl-tRNA synthetase familiy	1kmn	C	2.80	SYH_ECOLI
prolyl-tRNA synthetase	class-II aminoacyl-tRNA synthetase family	2i4o	A	2.40	SYP_RHOPA
Class II AARS homologue (bll0957)	class-II aminoacyl-tRNA synthetase family	3mey	A	2.50	Q89VT8_BRAJA
Lysyl-tRNA synthetase	class-II aminoacyl-tRNA synthetase family	3bju	A	2.31	SYK_HUMAN
glycyl-tRNA synthetase	class-II aminoacyl-tRNA synthetase family	2zt7	A	2.70	SYG_HUMAN
pyrrolysyl-tRNA synthetase	class-II aminoacyl-tRNA synthetase family	2q7g	A	1.90	PYLS_METMA
aspartyl-tRNA synthetase	class-II aminoacyl-tRNA synthetase family	3nem	B	1.89	SYD_PYRKO
Threonyl-tRNA synthetase	class-II aminoacyl-tRNA synthetase family	1nyr	A	2.80	SYT_STAAW
alanyl-tRNA synthetase	class-II aminoacyl-tRNA synthetase family	1yfr	A	2.15	SYA_AQUAE
serryl-tRNA synthetase	class-II aminoacyl-tRNA synthetase family	3lss	B	1.95	Q384V4_9TRYP
tRNA-lysidine synthase	tRNA(Ile)-lysidine synthase family	2e89	A	2.50	TILS_AQUAE
prolyl-tRNA synthetase	not classified	2j3m	B	2.30	Q831W7_ENTFA
serryl-tRNA synthetase	not classified	2cja	B	2.20	Q46AN5_METBA
NH3-dependent NAD+ synthetase	NAD synthetase family	1xng	B	1.70	NADE_HELPY
bacteriophage phi 6 RNA dependent RNA polymerase	Polymerase family	1hi1	A	3.00	RDRP_BPPH6
tRNA CCA-pyrophosphorylase	tRNA nucleotidyltransferase/poly(A) polymerase family	3h39	B	2.85	Q9WZH4_THEMA
polyA polymerase	tRNA nucleotidyltransferase/poly(A) polymerase family	3aqn	A	3.30	C9QS13_ECOD1
RNA editing ligase Mp52	not classified	1xdn	A	1.20	RLGM1_TRYBB
ATP hydrolysis, PPi is released and AMP is transferred (adenylation)
Protein Name	Family	PDB ID	chain	resol	Uniprot ID
nicotinamide mononucleotide (NMN) adenylyltransferase	archaeal NMN adenylyltransferase family	1f9a	A	2.00	NADM_METJA
phosphopantetheine adenylyltransferase	bacterial coaD family	1gn8	A	1.83	COAD_ECOLI
glucose-1-phosphate adenylyltransferase small	bacterial/plant glucose-1-phosphate	1yp3	C	2.60	GLGS_SOLTU
subunit	adenylyltransferase family
DNA polymerase IV	DNA polymerase type-Y family	3m9o	B	2.00	DPO42_SULSO
adenylyltransferase ThiF	hesA/moeB/thiF family	1zfn	A	2.75	THIF_ECOLI
lipoate-protein ligase A	lplA family	2aru	A	2.50	LPLA_THEAC
nicotinate-nucleotide adenylyltransferase	nadD family	1yun	A	2.00	NADD_PSEAE
pantoate-beta-alanine ligase	pantothenate synthetase family	2a84	A	1.55	PANC_MYCTU
polyA polymerase	poly(A) polymerase family	2q66	A	1.80	PAP_YEAST
tRNA CCA-pyrophosphrylase	tRNA nucleotidyltransferase/poly(A) polymerase family	3ovb	A	1.95	CCA_ARCFU
ubiquitin-activating enzyme E1C (Uba3)	ubiquitin-activating E1 family	1r4n	B	3.60	UBA3_HUMAN
ubiquitin-like 2 activating enzyme E1B	ubiquitin-activating E1 family	1y8q	D	2.25	ULE1B_HUMAN
ubiquitin-like modifier-activating enzyme 5	ubiquitin-activating E1 family	3h8v	A	2.00	UBA5_HUMAN
biotin protien ligase	not classified	2dto	A	1.50	O57883_PYRHO
FMN adenylyltransferase	not classified	3g59	A	1.87	Q6FNA9_CANGA
ATP hydrolysis, PPPi is relased and adenosine is transferred (adenosylation)
Protein Name	Family	PDB ID	chain	resol	Uniprot ID
methionine adenosyltransferase	AdoMet synthse family	1o9t	A	2.90	METK1_RAT
CoB(I)alamin adenosyltransferase	Cob(I)alamin adenosyltransferase family	1g5t	A	1.80	BTUR_SALTY
CoB(I)yrinic acid A,C-diamide adenosyltransferase	Cob(I)alamin adenosyltransferase family	2idx	A	2.50	MMAB_HUMAN
Others
Protein Name	Family	PDB ID	chain	resol	Uniprot ID
7,8-dihydro-6-hydroxymethylpterin-pyrophosphokinase	HPPK family	1dy3	A	2.00	HPPK_ECOLI
Preneck appendage protein	not classified	3gqn	A	2.15	B3VMP8_BPPH2
ATPsynthase epsilon subunit	ATPase epsilon chain family	2e5y	A	1.92	ATPE_BACP3
Eukaryotic peptide chain release factor subunit 1	eukaryotic release factor 1 family	3e1y	A	3.80	ERF1_HUMAN
prabable ATP-dependent RNA helicase Ddx58	helicase family	3lrr	A	2.15	DDX58_HUMAN
NAD-dependent malic enzyme	malic enzymes family	1gz4	A	2.20	MAOM_HUMAN
DCP2 protein	Nudix hydrolase family	2qkm	B	2.80	DCP2_SCHPO
acetylglutamate kinase-like protein	P(II) protein family	2rd5	D	2.51	GLNB_ARATH
STRADalpha	STE Ser/Thr protein kinase family	3gni	B	2.35	STRAA_HUMAN
redox-sensing transcriptional repressor Rex	transcriptional regulatory rex family	2vt3	B	2.00	REX_BACSU
transient receptor potential cation channel subfamily V member 1	transient receptor	2pnn	A	2.70	TRPV1_RAT
polyhedrin	not classified	2oh5	A	1.98	O10693_CPVBM
pertussis toxin subunit 4	not classified	1bcp	E	2.70	TOX4_BORPE
non-biological protein	not classified	2p09	A	1.65	—
5′-AMP-activated protein kinase catalytic subunit alpha-1	5′-AMP-activated protein kinase gamma subunit family	2v92	E	2.40	AAKG1_RAT
apoptosis regulator Ced4	AAA+ family/CARD domain/NB-ARC domain	2a5y	B	2.60	CED4_CAEEL
Clp1(inactive form)	Clp1 family	2npi	A	2.95	CLP1_YEAST
Rck dmain of YuaA protein	ktrA potassium transport (TC 2.A.38.4) family	2hmu	A	2.25	KTRA_BACSU
nitrogen regulatory protein P-II	P(II) protein family	2xbp	A	1.20	GLNB_SYNE7
O-sialoglycoprotien endopeptidase (probably miss annotation, in reality, AP endonuclease)	peptidase M22 family	2ivp	A	2.50	GCP_PYRAB
Rat synapsin I	synapsin family	1pk8	A	2.10	SYN1_RAT
putative uncharacterized protein TTHA0350	not classified	3ab8	A	1.70	Q5SLE3_THET8
phosphofruktokinase	not classified	3opy	B	3.05	Q8TGA0_PICPA
chloride channel protein 5 (clc-5)	chloride channel family	2j9l	C	2.30	CLCN5_HUMAN
gluconate kinase	FGGY kinase family	3ll3	A	2.00	Q5FM28_LACAC
Hypothetical protein YfbG	fmt family/ sugar epimerase familiy	1z7e	D	3.00	ARNA_ECOLI
ATP-dependent molecular chaperone Hsp82	heat shock protein 90 family	2cg9	B	3.10	HSP82_YEAST
DNA packaging protein Gp17	myoviridae large terminase family	2o0h	A	1.88	TERL_BPT4
AP4a hydrolase	Nudix hydrolase family	2pq1	A	1.95	O66548_AQUAE
aspartate carbamoyltransferase regulatory chain (PyrI)	PyrI family	2yww	B	2.00	PYRI_METJA
ribonucleotide reductase R1	ribonucleoside diphosphate reductase large chain family	3r1r	A	3.00	RIR1_ECOLI
SMC protein	SMC family	1xex	A	2.50	SMC_METJA
molybdenum storage protein subunit alpha	UMP kinase family	2ogx	A	1.60	MOSA_AZOVD
uncharacterized protein	universal stress protein A family	3cis	G	2.90	O06189_MYCTU
Actin-depolymerizing factor Brevin	villin/gelsolin family	2fgh	A	2.80	GELS_HORSE
ethanolamine utilization protein EutJ	not classified	3h1q	A	2.80	—
universal stress protein F	not classified	3fdx	A	1.58	A6T8F5_KLEP7
alcaligin biosynthesis protein	not classified	2x0q	A	1.96	P94255_BORBR
l-proline dehydrogenase alpha subunit	not classified	1y56	A	2.86	O59088_PYRHO
FtsA	not classified	1e4g	T	2.60	Q9WZU0_THEMA
NTRC-like two-domain protein	not classified	3fkq	A	2.10	—
hemerythrin-like domain protein DcrH	not classified	3a8t	A	2.37	Q9REU3_DESVU
Protein Mj1225	not classified	3lfz	A	2.20	Y1225_METJA
pyridoxal kinase	not classified	3ibq	A	2.00	Q88YB5_LACPL
ATP:CoB(I)alamin adenosyltransferase	not classified	2zhz	A	1.80	Q2SZ09_BURTA
clbalamin adenosyltransferase PduO-like protein	not classified	3gah	A	1.17	Q50EJ2_LACRE
putative ribokinase II	not classified	3iq0	B	1.79	Q8FD38_ECOL6
Universal stress protein family	not classified	2z08	A	1.55	Q5SJV7_THET8
phosphofruktokinase	not classified	3ie7	A	1.60	Q929S5_LISIN
achromobactin synthetase protein D (ACSD)	not classified	2x3j	A	2.00	Q93AT8_ERWCH
MccB	not classified	3h5n	A	1.90	Q47506_ECOLX
HipA	not classified	3dnt	B	1.66	HIPA_ECOLI
pyruvate carboxylase	not classified	3bg5	A	2.80	Q99UY8_STAAM
probable ATP-dependent DNA ligase D	not classified	2faq	A	1.90	Q9I1X7_PSEAE
ParA ATPase	not classified	3ea0	B	2.20	Q8KF94_CHLTE
small nucleolar RNP similar to Gar1	not classified	2hvy	B	2.30	Q8U029_PYRFU

We checked through the literatures of all these data for the biological function of ATP molecules and tabulated them based on the function. We found that 43 were for energy extraction through Pi hydrolysis, 42 for phosphorylation, 29 for energy extraction through PPi hydrolysis, 15 for adenylation, 3 for adenosylation and the remaining 56 were miscellaneous or function unknown (Table 1).

Molecular dynamics simulation of ATP in solvent

One of the results for 2 ns ATP simulations is shown in Figure 3. For the first 200 ps, the structure of ATP molecule seemed to oscillate amongst a limited number of conformations, but after that the molecule assumed many types of conformations. The behaviour in detail was different in different runs of simulation (Supplementary Figs. 1A–I), but the overall tendency and the scale of fluctuation were quite similar. For the analyses hereafter, we used all the conformations obtained in the latter 1 ns of ten runs, namely 100,000 samples of the conformations.

Figure 3

Root mean square deviation (RMSD) of ATP during the simulation. The calculation was done between the initial structure and structures of every 0.1 ps. All 36 atoms including hydrogen atoms were used in the calculation. Four snap shot structures were drawn in the graph. From left to right, conformations of 0 ps, 1,437 ps, 1,513 ps and 1,899 ps. This graph and the following ones were drawn by R21 except Figure 9.

Sufficiency of conformation sampling in this set of simulations is important in the following analyses. Figure 3 and Supplementary Figure 1 showed that, after 1 ns of simulation, ATP molecule underwent a compact and an extended conformations for a couple of times. These back-and-forth trajectories suggest that ATP molecule assumed quite a number of different conformations. In the following analyses, the analysis applied on conformations from each trajectory and the one applied to all as a whole did not show significant differences with a minor exception. This behaviour of the data suggests that the reasonable number of conformations was obtained in the ten runs of 2 ns simulation.

Comparison of ribose conformations

Curvature of ribose in ATP had different distributions between the one calculated from the snapshot conformation in MD simulation and the one from PDB data (Fig. 4). The Gaussian curvature of the ribose from MD simulation had normal-like distribution around −0.11 and the mean curvature had normal-like distribution around 0.02. This behaviour was almost the same in each trajectory of ten runs (Supplementary Fig. 2). The distribution of the mean curvature of the ribose from PDB was more or less the same as the distribution from MD simulation, but the distribution of the Gaussian curvature of the ribose from PDB was not in the normal form and about 70% of them lay between −0.10 and −0.05. The value of the Gaussian curvature is always negative by definition, and when the value is close to zero, the ring structure is close to a flat structure. Therefore, the comparison of the structures above suggests that the ribose in ATP is off the plane when it exists in water, but is restricted to relatively planar structure when bound to a protein. This difference is not that obvious when the structures are compared in torsion angles of the ribose ring.

Figure 4

Ribose curvature in the conformations from molecular dynamics simulation and from PDB. A black dot is obtained from the snap shot conformation form the molecular dynamics simulation, and a red dot is from PDB. The histogram in black clarifies the distribution of black dots, and the one in red clarifies the distribution of red dots. The ribose with minimum/maximum curvature values in the snap shot conformations from the molecular dynamics simulation were drawn on the histograms.

The torsion angles τ0 and τ4 can be good indicators of puckering structure of ribose ring. As shown in Figure 5, a cluster of structures at the first quadrant (τ0>0 and τ4>0) is C2′ exo conformation, the second quadrant (τ0<0 and τ4>0) is O4′ endo conformation, the third quadrant (τ0<0 and τ4<0) is basically C2′ endo conformation, and the fourth quadrant (τ0>0 and τ4<0) is O4′ exo conformation. In water, C2′ exo and C2′ endo conformations was highly dominated followed by O4′ endo conformation. When the distribution in different ten runs of simulation was examined (Supplementary Fig. 3), four runs (trajectories 01, 03, 06, 07) had more numbers of C2′ exo conformations and two runs (trajectories 05, 09) had more numbers of C2′ endo conformations. As a whole, there is a tendency to prefer both C2′ exo and C2′ endo conformations in water. When ATP bound to protein, the number of C2′ exo and C2′ endo conformations were more or less the same and O4′ endo conformation was less populated.

Figure 5

Ribose torsion angles τ0 and τ4 in the conformations from molecular dynamics simulation and from PDB. A black dot is obtained from the conformation of molecular dynamics simulation, and a red dot is from PDB. The histogram in black clarifies the distribution of black dots, and the one in red clarifies the distribution of red dots.

The difference in puckering seemingly has a connection to the biological role of ATP molecules. Out of 188 protein-bound ATP molecules in the dataset, 43 ATP molecules were for energy extraction through Pi hydrolysis, and 42 ATP molecules were for phosphorylation (Table 1). About 50% of 43 plus 42 ATP molecules took either C2′ exo or C2′ endo conformation. Interestingly, 33% of ATP molecules in energy extraction group (the maximum portion in the group) took C2′ endo conformation, and 33% of ATP molecules in phosphorylation group (the maximum portion in the group) took C2′ exo conformation.

Comparison of adenine conformation

We analyzed the conformation of adenine in two separate rings, namely five-membered ring and six-membered ring. The five-membered ring had a flat conformation during the MD simulation with an occasional slight deviation (Fig. 6). The distribution of the black dots in the figure, which forms an eastbound comet shape in any runs of simulation (Supplementary Fig. 4), suggests that the five-membered ring in adenine should undergo puckering in a very slight scale. The five-membered rings of adenine in the ATP molecules in PDB took a very flat conformation as visualized in the figure by red dots. Almost all the dots were found at the head of the comet shape, where both Gaussian and mean curvatures were very close to zero.

Figure 6

Adenine five-membered ring curvature in the conformations from the molecular dynamics simulation and from PDB. A black dot is obtained from the conformation of the molecular dynamics simulation, and a red dot is from PDB. The histogram in black clarifies the distribution of black dots, and the one in red clarifies the distribution of red dots. The adenine five-membered rings with minimum/maximum curvature values in the snap shot conformations from the molecular dynamics simulation were drawn on the histograms. A chemical bond at the bottom of each figure is a glycosyl bond and six-membered ring is located at the far side.

The conformation of six-membered ring in adenine had different characteristics compared with the five-membered ring. In the conformation obtained by the MD simulation, the distribution of the Gaussian curvature was significantly different from that for the five-membered ring (Fig. 7). In the Gaussian curvature, the absolute value of the center of the distribution was significantly greater, and the width of the distribution was significantly wider than those of five-membered ring. The magnitude of distribution in the mean curvature was also greater than that of five-membered ring. These differences evidently appeared in any runs of the simulations (Supplementary Fig. 5). All of these facts indicate that the six-membered ring in solution was deviated from a flat structure in a greater scale compared with the five-membered ring. These deviations from flatness were, however, considerably adjusted when ATP molecule bound to a protein. The distribution of Gaussian curvature of six-membered ring in PDB protruded out to the east direction from the distribution of the Gaussian curvature and squeezed to the center of the mean curvature of ATP in water (red dots in Fig. 7). The six-membered ring of adenine was apparently flattened by the protein, to the extent of the flatness that rarely appeared in ATP in water.

Figure 7

Adenine six-membered ring curvature in the conformations from the molecular dynamics simulation and from PDB. A black dot is obtained from the conformation of the molecular dynamics simulation, and a red dot is from PDB. The histogram in black clarifies the distribution of black dots, and the one in red clarifies the distribution of red dots. The adenine six-membered rings with minimum/maximum curvature values in the snap shot conformations from the molecular dynamics simulation were drawn on the histograms except for the conformation on the far right side which is derived from PDB structure (PDB ID: 2J9L). A chemical bond at the bottom of each figure is a glycosyl bond and five-membered ring is located at the far side.

Adenine structure can be approximated to two flat rings that oscillate at the connection and the oscillation motion can be observed in the MD simulation. We described the oscillation motion by defining a pseudo-ring across the two rings and calculated Gaussian and mean curvatures (Fig. 8). In the conformation obtained from the MD simulation, both Gaussian and mean curvatures had normal-like distribution and a crescent-shape distribution when combined; two edges of the crescent consisted of the conformations in the long tail of the Gaussian curvature. These distributions were observed in trajectories of ten runs (Supplementary Fig. 6). In the conformations from PDB, however, the values of the mean curvature were virtually zero and the values of the Gaussian curvature distributed around two peaks, namely the peaks at −0.75 and at −0.68. The former conformations mostly lay within the distribution of ATP in solution, but the latter conformations lay out of the range of the distribution of ATP in solution. The distribution of Gaussian curvature in PDB had no clear correlation to other values such as buriedness of ATP molecule to the protein or the function of ATP molecules, and hence the physicochemical explanation for this distinction needs further study. It seems that, due to some structural constraints, the conformation with Gaussian curvature −0.70 is prohibited in the adenine ring.

Figure 8

Adenine hinge motion. The hinge motion is defined by the open-book movement in five-membered and six-membered rings in the adenine molecule. A pseudo-ring was defined to assess the openness of the hinge. See the method section for the detail. A black dot is obtained from the conformation of the molecular dynamics simulation, and a red dot is from PDB. The histogram in black clarifies the distribution of black dots and the one in red clarifies the distribution of red dots. The hinge conformations with minimum/maximum curvature values in the molecular dynamics simulation were drawn. A chemical bond at the bottom of each figure is a glycosyl bond and six-membered ring is located at the far side.

Different distributions of torsion angles between the conformations of MD simulation and of PDB

The torsion around the chemical bond between the phosphate unit and the ribose (γ), and that around the glycosyl bond connecting the ribose and adenine (χ) are apparently far more flexible than the torsion angles around the bonds for ribose and adenine rings in ATP molecule (Fig. 1). However, the torsion angles around these bonds in conformations from MD simulation were heavily populated at only two states. When the conformations were counted with the bins of torsion angles digitized by 10 degrees, the densely populated bins were represented by a pair of torsion angles γ=−170 and χ=70, and by a pair of γ=−60 and χ=60. Both conformations were found around 1.0% of the whole population (Fig. 9). Different trajectories had peak population in different torsion angle pairs (Supplementary Fig. 7), but the two peaks in Figure 9 were almost consistently appeared as one of the top peaks in all trajectories. The noticeable exceptions were trajectories 5 and 6. Both trajectories did have a peak at γ=−60 and χ=60, but did not have a peak at γ=−170 and χ=70. The torsion angles γ=−90 to −180 represents a trans conformation between O5′ and C3′. The torsion angle χ=60 represents a gauche+ or syn conformation between the ribose and the adenine. Obviously the ATP molecule assumes a compact conformation by syn conformer in water.

Figure 9

Probability density function map of the torsion angles &gamma and χ. The left map is derived from the snap shot conformations of the molecular dynamics simulation, and the right map is from the conformations in PDB. The probability is depicted in rainbow colour scheme from blue to red in ascending order as shown in the colour bars. Note that the dynamic range of the two maps is different. One of the structures in highly populated torsion angles is shown on the top.

Peaks in a pair of torsion angles were found in different values in the conformations from PDB. The most heavily populated pair of angles was γ=50, χ=−150 (3.4%), followed by γ=50, χ=−160 (2.9%) and γ=40, χ=−120 (2.9%) (Fig. 9). The torsion angle γ=50 represents a cis conformation between O5′ and C3′.χ=−120 to −160 represents an anti conformation between the ribose and the adenine. When bound to a protein, the ATP molecule is extended over the protein.

In the population derived from MD simulation, the proportion of the conformations abundant in PDB was approximately half of the most populated conformation. Both the conformations with γ=50 and χ=−150 and the conformations with γ=50 and χ=−160 occupied about 0.4%, and the conformations with γ=40 and χ=−120 about 0.2%. In trajectory 6 in ten runs of simulations, 1% of the population was found in a pair of torsion angles close to the conformations found in PDB. This is, however, the only run with the dense population and none of the nine others had the dense population at the corresponding torsion angle pairs. On the other hand, in the population of PDB, the proportion of the conformations abundant in MD simulation was virtually none. These results strongly suggest that during the process of ATP binding to protein, the protein should exert forces on ATP molecule to assume the specific conformation that were under-represented in solution.

As mentioned above, there were three sets of torsion angles in ATP molecules that often appeared in PDB. These three sets were virtually grouped into two, namely, a pair of 50≤γ<60 and −160<χ≤−140, and a pair of 40≤γ<50 and −120<χ ≤−110 (Fig. 9). When we examined the function of ATP molecules in both peaks, we found that the proteins in the former peak had ATP for phosphorylation function twice as many as those in the latter peak (the second group in Table 1). Mildvan discussed in his review17 and his works with the coworkers, that the former peak of χ angle (they called low-antiglycosyl torsional angle) was found in ATP-Mn2+ binary complex and represented presumably an inactive form, and that the latter peak of χ angle (they called high-antiglycosyl torsional angle) was found in ATP-Mn2+-kinase ternary complex and presumably represented an active form. Combined with the current analyses, we suggest that the former peak (50≤γ<60 and −160<χ≤−140) is the set of torsion angles for inactive form and may be easily crystalized. And the latter (40≤γ<50 and −120<χ≤−110) peak is the torsion angles for active form and may be difficult for crystalization, because the conformation initiates chemical reactions. This may explain the difference in the density of population in two peaks. The over-representation of ATP molecules for phosphorylation in the former peaks can be explained by the possibility that they were much easily crystalized in the inactive form.

Conclusion

In this paper, we extensively analyzed the effect of protein to ATP conformations. It has been implicitly assumed that protein affects on ATP conformation when it binds, but there were no comprehensive study on this issue.

Based on the sampling of the ATP solution structures obtained from MD simulation, and the sampling of ATP structures bound to a protein in Protein Data Bank, the following three characteristics were found.

The ribose ring in ATP molecule, which is flexible in solution, tends to assume C2′ exo or C2′ endo conformation when it binds to protein. Proteins that use ATP for energy source tend to bind ATP with C2′ endo forms. Proteins that use ATP for phosphorylation tend to bind ATP with C2′ exo forms.

The adenine ring in ATP molecule, which assumes open-book motion with the two ring structures, has two distinct structures when ATP binds to protein. One of the structures is commonly found in solution but the other not. The physicochemical background of this distinction needs further study.

The torsion angles of glycosyl bond (χ) and the bond between phosphate unit and the ribose (γ) take unique values when ATP binds to protein. The combination of the torsion angles well populated in solution rarely found in the ATP molecule on the protein. There are two well-populated torsion angles in ATP bound to proteins, one of which may represent active form and the other inactive form.

These findings suggest that ATP-binding protein forces ATP to take rare conformation in solution when ATP binds to protein, and that this conformational change exerted by the protein should be the trigger for the cleavage of the γ phosphate group.

Finding a conformation of the bound ligand is a big issue in protein-ligand docking problem18,19,20. The widely used methods introduced MD to search for the conformation of the ligand placed close to the protein. The current study implies that, in the case of ATP molecule, protein bound conformation can hardly be achieved by simple MD simulation, as shown that flatness of the ring structures and the χ and γ torsion angles for protein-bound ATP rarely appears in solution. Therefore, a sophisticated MD simulation that includes both a ligand and a protein at once is, at least, necessary to sample the conformations for protein-ligand complex. In addition, the failure in finding the appropriate conformation in MD simulation can be circumvented by a database search (database sampling), in case the protein-ligand conformations are abundant in the database.