RESEARCH ARTICLE
`
`Structures of Ternary Complexes
`of Rat DNA Polymerase , a DNA
`Template-Primer, and ddCTP
`Huguette Pelletier, Michael R. Sawaya, Amalendra Kumar,
`Samuel H. Wilson, Joseph Kraut
`Two ternary complexes of rat DNA polymerase p (pol ,), a DNA template-primer, and dide-
`oxycytidine triphosphate (ddCTP) have been determined at 2.9 A and 3.6 A resolution,
`respectively. ddCTP is the triphosphate of dideoxycytidine (ddC), a nucleoside analog that
`targets the reverse transcriptase of human immunodeficiency virus (HIV) and is at present
`used to treat AIDS. Although crystals of the two complexes belong to different space groups,
`the structures are similar, suggesting that the polymerase-DNA-ddCTP interactions are not
`affected by crystal packing forces. In the pol p active site, the attacking 3'-OH of the
`elongating primer, the ddCTP phosphates, and two Mg2+ ions are all clustered around
`Asp19g, Asp192, and Asp256. Two of these residues, Asp190 and Asp256, are present in the
`amino acid sequences of all polymerases so far studied and are also spatially similar in the
`four polymerases-the Kienow fragment of Escherichia coli DNA polymerase 1, HIV-1
`reverse transcriptase, T7 RNA polymerase, and rat DNA pol 1-whose crystal structures are
`now known. A two-metal ion mechanism is described for the nucleotidyl transfer reaction and
`may apply to all polymerases. In the ternary complex structures analyzed, pol
`binds to the
`DNA template-primer in a different manner from that recently proposed for other polymerase-
`DNA models.
`
`their cellular dNTP counterparts by the
`absence of an attacking 3'-hydroxyl group
`(3'-OH) (Fig. 2) and therefore, once a
`dideoxynucleotide is successfully incorpo-
`rated into a growing primer strand, there
`can be no further incorporation of subse-
`quent nucleotides. A well-known example
`of this kind of inhibition involves HIV-1
`reverse transcriptase (RT), which is the
`polymerase responsible for the replication of
`the HIV genome. 3'-azido-2',3'-dideoxy-
`thymidine
`(AZT),
`2',3'-dideoxyinosine
`
`DNA replication (1) is a highly complex
`biological process, even for a relatively
`simple organism such as Escherichia coli.
`During replication,
`the double
`helical
`DNA molecule is unwound, and the two
`resultant single strands of DNA act as
`templates to guide the synthesis, one com-
`plementary base at a time, of antiparallel
`primer strands. Although many auxiliary
`ligases,
`proteins such as
`helicases, and
`topoisomerases are usually involved, the
`chemical reaction at the core of DNA
`replication, the nucleotidyl transfer reac-
`tion, is catalyzed by DNA polymerases and
`may be depicted as follows:
`Template-primer, + dNTP
`template-primer-dNMPn,1 + PPi
`where dNTP (2'-deoxyribonucleoside 5'-
`triphosphate) represents any one of four de-
`oxynucleotides (dATP, dGTP, dCTP, and
`dTTP), and dNMP and PPi represent 2'-
`deoxyribonucleoside 5'-monophosphate and
`pyrophosphate, respectively (Fig. 1).
`Inhibition of a polymerase that effects
`genomic replication can be fatal to an
`organism. In a common type of polymerase
`inhibition,
`2',3'-dideoxynucleotides
`(dd-
`NTPs) act as chain terminators of the
`primer strand. The ddNTPs differ from
`
`H. Pelletier, M. R. Sawaya, and J. Kraut are in the
`Department of Chemistry, University of California, San
`Diego, CA 92093-0317, USA. A. Kumar and S. H.
`Wilson are at the Sealy Center for Molecular Science,
`University of Texas Medical Branch, Galveston, TX
`77555-1051, USA.
`
`(ddI), and 2',3'-dideoxycytidine (ddC) are
`all anti-HIV drugs (2, 3) that become potent
`chain termination inhibitors of RT after
`they are converted by cellular kinases (4, 5),
`in vivo, to their corresponding nucleoside
`5'-triphosphates, AZT-TP, ddATP (6), and
`ddCTP, respectively. In that all polymerases
`probably share a common catalytic mecha-
`nism, it is not surprising that some toxic
`effects of these drugs have been attributed to
`inhibition of host-cell polymerases, perhaps
`including the pol 1B described here (7-9).
`Therefore, a detailed understanding of the
`nucleotidyl transfer reaction, as well as the
`mechanism of inhibition of viral and host
`cell polymerases by nucleoside analogs, may
`lead to the design of more potent and less
`toxic HIV-1 RT inhibitors for use in the
`treatment of AIDS.
`Despite limited sequence similarity to
`the Klenow fragment (KF) of E. coli DNA
`pol I [the only other polymerase for which a
`crystal structure (10) was known at the
`time], the crystal structure determinations
`of HIV-1 RT (11, 12) revealed a common
`polymerase fold consisting of three distinct
`subdomains (designated fingers, palm, and
`thumb because of the resemblance to a
`hand) forming an obvious DNA binding
`channel. The strongest structural overlap
`between KF and RT comprised a trio of
`carboxylic acid residues located in the palm
`subdomain (11, 12). These observations led
`to the hypothesis that perhaps all polymer-
`ases share a common nucleotidyl transfer
`mechanism centered around the highly
`conserved carboxylic acid residues
`(11).
`Strengthening this argument somewhat was
`the subsequent crystal structure determina-
`tion of an RNA polymerase (RNAP) from
`bacteriophage T7, which showed strong
`structural similarities with KF (13). How-
`
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
` on March 5, 2014
`
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`www.sciencemag.org
`
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`Downloaded from
`
`Pdimer
`
`I
`
`5
`
`Pfimer
`0U
`5'
`o--o 5
`0
`
`Base
`
`'° X
`
`'J
`
`3
`
`HO-P-O-P-OH + B-P-0-
`0
`0
`
`dNMPn
`'
`
`Base
`
`dNMPn
`
`Base
`
`dNTP
`
`Pp.
`
`dNMPn+l
`
`OH
`Fig. 1. The nucleotidyl transfer reaction. The 3'-OH group of the terminal dNMP on the primer strand
`attacks the 5'-a phosphate of an incoming dNTP, and a newly formed phosphodiester linkage
`results in elongation of the primer strand by one dNMP. After release of pyrophosphate (PP), the
`catalytic cycle is complete and the 3'-OH group of the newly incorporated dNMP is now ready to
`attack yet another incoming dNTP. Only the 3' end of a primer is extended so that DNA
`polymerization is said to proceed in a 5' to 3' direction. If the polymerase molecule does not release
`the template-primer before incorporation of a second dNMP, the mode of DNA synthesis is said to
`be "processive", but if the polymerase releases the template-primer after each successive
`incorporation of a dNMP, the mode of DNA synthesis is said to be "distributive".
`
`SCIENCE * VOL. 264
`
`*
`
`24 JUNE 1994
`
`1891
`
`Illumina Ex. 1044
`IPR Petition - USP 10,435,742
`
`

`

`iM1i 0
`
`il
`
`----------------
`
`~.
`
`4JA-ddCTP sample was prepared at room
`nperature; approximately 1.2 mg of the
`-nucleotide (nt) template and 0.8 mg of
`c 6-nt primer (27) were dissolved in 240
`of a buffer solution (20 mM MgCI2, 0.1
`MES, pH 6.5) and the mixture was left
`a sealed Eppendorf tube for 1 hour to
`ow annealing of the template-primer
`(240-pl of a
`3). Two portions of pol 1
`lution containing 20 mg/ml) were then
`awed and added, and the protein-tem-
`te-primer sample was allowed to stand
`an additional hour. A 4-pgmol sample of
`CTP (in 40 ,ul of H20) (29) was the last
`mponent to be added, resulting in a
`iction mixture containing pol 1
`at ap-
`Dximately 10 mg/ml, 10 MM MgC12, and
`excess of template:primer:ddCTP in mo-
`ratios of 3:4:30, respectively, relative to
`c amount of protein. The reaction, nu-
`otidyl transfer of ddCMP to the primer
`terminus, was allowed to proceed for 2
`urs before crystallization trays were set up
`
`Two different crystal forms were ob-
`ved, depending on the concentration of
`hium sulfate in the reservoir solution.
`ie crystal form, obtained with lithium
`Lfate concentrations from 40 to 75 mM,
`IS hexagonal and grew to dimensions of
`3 by 0.8 by 0.6 mm in about 2 weeks.
`ese crystals belong to space group P61
`= b = 94.9, c = 117.6 A), with one
`mplex molecule per asymmetric unit.
`ider similar conditions, but at lithium
`Lfate concentrations from 75 to 150 mM,
`telike crystals grew to dimensions of
`
`O
`
`sN
`
`H
`
`H
`
`Fig. 2. Two normal cellular
`nucleotides, (A) 2'-deoxy-
`cytidine
`5'-triphosphate
`(dCTP) and (B) 2'-deoxy-
`thymidine 5'-triphosphate
`(dTTP), and their anti-HIV
`(C)
`(drug)
`counterparts
`0 2',3'-dideoxycytidine
`5'-
`triphosphate (ddCTP) and
`(D) 3'-azido-2',3'-dideoxy-
`thymidine 5'-triphosphate
`(AZT-TP). The triphosphate
`moiety, which is linked via
`a phosphoester bond to
`the 5' carbon of the ribose,
`is designated TP.
`
`ever, because structural evidence for a com-
`mon nucleotidyl transfer mechanism has so
`far been limited to comparisons among
`polymerases from a bacterium (KF), a virus
`(RT), and a phage (RNAP), perhaps the
`most convincing evidence for this hypoth-
`esis is provided by the crystal structure
`determination of a eukaryotic polymerase,
`rat DNA pol 13(14). Sequence alignments
`show that pol 13 is so distantly related, even
`from its eukaryotic relatives, polymerases a,
`y, 8, and e, that it stands in a class of
`its own along with only one other polymer-
`ase, terminal deoxynucleotidyltransferase
`(TdT) (15). The crystal structure of pol P
`nevertheless revealed a polymerase fold
`consisting of palm, fingers, and thumb
`(along with an additional 8-kD domain
`attached to the fingers), and the most
`striking structural similarity with its distant
`relatives, KF, RT, and RNAP, is a portion
`of the palm that bears the highly conserved
`carboxylic acid residues (14). This suggests
`that despite the large differences in size (pol
`13, at 39 kD, is the smallest polymerase
`known), in function [although pol 13 may
`play a role in DNA replication (16, 17), its
`primary function is in DNA repair (18-
`20)], and in fidelity [pol 13 is the most error
`prone eukaryotic polymerase studied to date
`(21, 22)], pol 13 probably shares a common
`nucleotidyl transfer catalytic mechanism
`with all other polymerases.
`Taking advantage of the chain termina-
`tion method of polymerase inhibition with
`ddNTPs, we have succeeded in growing
`crystals of rat pol 13 complexed with two
`pseudo substrates, namely, (i) a DNA tem-
`plate-primer in which the 3' end of the
`primer has been "terminated" by ddCMP,
`and (ii) ddCTP. In preparation for crystal-
`lization experiments, we mixed pol 13 with
`the DNA template-primer shown below
`and a large excess of ddCTP on the assump-
`tion that, prior to crystallization, the fol-
`lowing reaction would occur:
`3'GCCGCGGGAAA5'
`3' + ddCTP --
`5'CGGCGC
`3'GCCGCGGGAAA5'
`5'CGGCGCCdd
`3' + PPi
`the newly incorporated
`where Cdd is
`ddCMP. If pol 1
`then were to try to
`incorporate another nucleotide onto the
`a second nucleotidyl
`primer terminus,
`transfer reaction could not occur because
`the recently incorporated ddCMP lacks a
`3'-OH group. This should result in a pseudo
`complex
`Michaelis-Menten
`ternary
`in
`which both "substrates" are present, name-
`ly, a nonreactive template-primer and a
`nucleoside triphosphate. Crystals were ob-
`tained, and the subsequent structure deter-
`minations revealed that this must have been
`what happened. Electron density maps
`
`showed a primer strand that was seven nu-
`cleotides long, although we started with a
`primer that was only six nucleotides in
`length, and the 3' (deoxy) terminus of the
`primer was positioned next to strong elec-
`tron density resembling a nucleoside triphos-
`phate, presumably ddCTP.
`Such a detailed view of the active site in
`the ternary complex allows us to propose a
`two-metal ion mechanism for the nucleoti-
`dyl transfer reaction that is similar, in many
`ways, to the two-metal ion mechanism
`previously proposed for another type of
`phosphoryl transfer reaction-the exonu-
`clease reaction of the 3Y'-5' exonuclease of
`E. coli DNA pol I (23, 24). Our proposed
`nucleotidyl transfer mechanism probably
`applies to all polymerases, but when we
`attempt to extend that mechanism to the
`other three polymerases-KF, RT, and
`RNAP-for which the crystal structures are
`known, a problem arises: in our structures,
`pol 13 is bound to the DNA in a manner
`that differs from the recently proposed poly-
`merase-DNA models for all three of these
`polymerases.
`Crystailizations and preliminary dif-
`fraction studies. Recombinant rat DNA
`pol 13 (25) was expressed in E. coli and
`purified as described (26). After purifica-
`tion, the protein was washed three times in
`a microconcentrator (Centricon- 10, Ami-
`con) with a buffer solution (10 mM ammo-
`nium sulfate, 0.1 M tris, pH 7.0), then
`concentrated to 20 mg/ml and stored at
`-80°C in sealed Eppendorf tubes (120-,ul
`portions). Prior to crystallization, a protein-
`
`A
`
`B
`
`dCTP
`
`dTTP
`
`C
`
`D
`
`ddCTP
`
`N3
`AZr-TP
`
`1 892
`
`SCIENCE * VOL. 264
`
`*
`
`24 JUNE 1994
`
`

`

`1.0 by 0.6 by 0.2 mm in a few days. These
`(a
`crystals belong to space group P21
`= 106.3, b = 56.8, c = 86.7 A, and 1 =
`106.4°) and there are two complex mole-
`cules per asymmetric unit. Often both crys-
`tal forms grew in the same drop, and the
`crystals on which data were collected (Ta-
`ble 1) were grown at the same concentra-
`tion of lithium sulfate, 75 mM. The unusu-
`ally large excess of ddCTP (1:30 molar
`ratio) was required in order to obtain the
`P61 crystals, but the P2, crystals could be
`grown under much lower ddCTP excesses
`(1:10 molar ratio). Extreme purity of all
`components in the crystallization medium,
`particularly the DNA samples (27), seemed
`to be an absolute requirement for growing
`both types of temary complex crystals.
`Attempts were made to obtain temary
`complex crystals of rat pol ,B, a DNA tem-
`plate-primer, and AZT-TP (Fig. 2D) (31)
`under similar conditions, even though incor-
`poration of AZT-MP would result in a mis-
`matched base pair (of a G-T type) at the
`primer terminus (22). Orthorhombic crystals
`grew in space group P21212 (a = 188.4, b =
`67.7, c = 39.1 A) with one pol P molecule
`in the asymmetric unit. A 4.0 A data set was
`collected and preliminary structural studies
`(32) showed that, because of crystal pack-
`ing, it was not possible for the template-
`primer to occupy the pol 13 binding channel.
`Failure of pol 13 to form a tight complex with
`the DNA template-primer under these con-
`ditions might be attributed to steric hin-
`drance by the azido group of a newly incor-
`porated AZT-MP on the primer terminus.
`Efforts to obtain a binary complex of pol
`1 and a DNA template-primer alone (nei-
`ther ddCTP nor AZT-TP) resulted in crys-
`tals that grew under much different condi-
`
`.SR
`
`.X~s~w> f~f'S~. m
`
`tions, but were nonetheless isomorphous
`with the P21212 (AZT-TP) crystals men-
`tioned above. Failure of pol 13 to bind to the
`DNA in this case could be due to the higher
`salt concentration of the crystallization me-
`dium (about 250 mM salt compared to 75
`mM). Because one crystallization medium
`contained AZT-TP and the other did not,
`we calculated FO(AZT-TP) -FO(apo) oLC, dif-
`ference Fourier maps to see whether an
`AZT-TP binding site could be located.
`Strong electron density was observed in an
`area of the map adjacent to Arg'49, which is
`near the catalytically important residues,
`Asp'92, and Asp256. This pol
`Asp'90,
`1-AZT-TP binary complex which, as dis-
`cussed below, is probably not catalytically
`relevant, is similar to a KF-dNTP binary
`complex in which the dNTP bound to Arg
`residues near the catalytically important
`carboxylic acid residues of KF (33).
`Human pol 1, which has been cloned
`and expressed (34, 35) in a manner similar
`to that of rat pol 1, shares more than 95
`percent sequence similarity with rat pol 1,.
`so it was somewhat surprising when at-
`tempts to obtain ternary complex crystals
`of human pol 1, a DNA template-primer,
`and ddCTP under the same conditions de-
`scribed above for the rat enzyme resulted
`in crystals that grew in two previously
`unobserved orthorhombic crystal forms.
`One form has unit cell parameters a =
`158, b = 108, c = 60 A, with probably
`two complex molecules in the asymmetric
`unit, but the crystals diffract only to about 5
`A resolution. In contrast, the second crystal
`form diffracts quite well (to about 3.3 A),
`but its unit cell parameters of a = 465, b =
`168, c = 56 A are so large that special data
`collection techniques would be required.
`
`Table 1. Data collection and refinement statistics. X-ray diffraction data were collected on a
`multiwire area detector (98) (San Diego Multiwire Systems) with monochromatized CuKa radiation
`(Rigaku rotating anode x-ray generator), and intensity observations for each data set were
`processed with a local UCSD Data Collection Facility software package (99). Reflections from 20 A
`to the maximum resolution were included in all least squares refinement steps. The final structures
`for both complexes include all residues, with the exception of residues 1 to 8 of the disordered
`NH2-terminus and residues 246 to 248 of a disordered surface loop. There are a few missing side
`chain atoms in both coordinate sets that are mainly in lysine and arginine residues of the 8-kD
`domain. Omit maps were used to confirm the modeling of the DNA template-primer, the ddCTP
`nucleotide, and the cis-peptide bond between Gly274 and Ser275.
`
`Data collection
`
`Refinement
`
`Space
`group
`
`dmin
`(A)
`
`a *
`
`Reflections
`
`rotal
`
`Unique
`
`rms
`deviation§
`Bond
`Angle
`(0)
`(A)
`P61
`0.193
`2.9
`0.020
`0.087
`2914
`53,583
`13,281
`99
`1.8
`2.9
`P21
`0.199
`0.018
`2.9
`5753
`0.059
`25,046
`3.6
`10,650
`96
`1.8
`tRSym = lI'obs
`*Average ratio of observed intensity to background in the highest resolution shell of reflections.
`*The number of nonhydrogen atoms includes 31 and 4 water oxygens for the P61 and the P21
`/avg.
`-
`strucures, respectively.
`§The rms bond and rms angle values are the deviations from ideal values of the bond
`IlFinal R = SIFObS - FcacI/1Fobs, including all data between 20 A
`lengths and bond angles in the final model.
`and the maximum resolution.
`
`Complete-
`ness (%)
`
`Y
`
`Final
`R11
`
`SCIENCE * VOL. 264
`
`*
`
`24 JUNE 1994
`
`Structure determination and refine-
`ment. Data collection and refinement sta-
`tistics for the structure determinations of
`the two ternary complexes of rat pol 1, a
`DNA template-primer, and ddCTP are list-
`ed in Table 1. Structure solutions utilized
`the refined atomic coordinates of the high
`resolution (2.3 A) structure of the 31-kD
`domain of rat pol 1
`(14). The molecular
`replacement programs of XPLOR (36) gave
`clear rotation solutions for the 31 -kD do-
`main of both ternary complexes, but only
`after results from classical cross-rotation
`searches had been filtered through the Pat-
`terson-correlation (PC) refinement steps
`(37). PC refinement techniques were par-
`ticularly powerful for our structure determi-
`nations because independent rigid body
`movements of the fingers, palm, and thumb
`subdomains of the 31-kD domain could be
`allowed during PC refinement of the rota-
`Results from subsequent
`searches.
`tion
`translation searches gave solutions for the
`P61 complex structure that were, in gener-
`al, at higher peak height to background
`ratios than translation solutions for the P21
`complex structure, but the highest transla-
`tion peaks in both cases nevertheless were
`the correct solutions (38).
`The 31-kD partial structure solutions
`obtained by molecular replacement tech-
`niques were subjected to rigid body refine-
`ment by XPLOR (36), where the entire
`31 -kD domain was first allowed to move as
`a rigid body, then later, the fingers, palm,
`and thumb subdomains were allowed to
`move as independent rigid bodies simulta-
`neously. Typical R factors at this stage were
`about 50 percent. After subsequent posi-
`tional refinement with the least squares
`program package TNT (39) had lowered
`the R factors of the partial solutions to
`about 45 percent, we calculated Fo - FC,
`a, difference Fourier maps that revealed
`clear electron density for many of the back-
`bone phosphates of a double-stranded DNA
`molecule as well as the three phosphates of
`a ddCTP nucleotide, and even portions of
`the 8-kD domain were evident at this early
`stage. Cycles of model building and least
`squares refinement improved the electron
`density for the rest of the DNA as well as
`the 8-kD domain for both complex struc-
`tures, and once the R factors had dropped
`below 30 percent, refinement of individual
`isotropic temperature factors also improved
`the maps and facilitated refinement.
`Although we were unable to discern the
`DNA base sequences at these resolutions,
`the directionality (5' -
`3') of the DNA
`strands was evident early in our modeling
`efforts, hence we knew that the 3' terminus
`of either the template strand or the primer
`strand was positioned at the pol 13 active
`site. What ultimately distinguished the
`template from the primer was that we were
`1893
`
`

`

`loll
`
`., 1.
`
`.ll
`
`gowicamlel
`
`.ll
`
`1110011
`
`9 1 INWORM.11
`
`able to model in seven nucleotides for one
`of the DNA strands and at least 8 nt for the
`second DNA strand. Provided that no un-
`expected side reactions had occurred during
`crystallization, we knew that the primer
`could be no longer than 7 nt, and therefore
`the DNA strand containing 8 nt was desig-
`nated the template. We then concluded
`that the first three bases of the template
`(AAA) are disordered in both crystal struc-
`tures. This interpretation of the data is in
`agreement with the idea that the 3' termi-
`nus of the primer should be positioned at
`the polymerase active site. Analysis of the
`DNA in our refined structures with the
`program CURVES II (40) indicated that
`the DNA is predominantly B form. Our
`DNA may have some A-DNA characteris-
`tics, however, in that the minor groove
`width appears to increase as the DNA
`approaches the pol 13 active site; the section
`of the double-stranded DNA that is re-
`moved from the active site and protrudes
`into solution is characteristic of B-DNA
`with a minor groove width of 11 A, whereas
`nearer to the active site, the minor groove
`width is almost 15 A (typical A-DNA has a
`minor groove width of about 17 A).
`Description of the structures. When
`the pol P3 ternary complex structures are
`compared with the structure of the apo
`enzyme (Fig. 3), the most apparent differ-
`ences consist of large movements of the
`8-kD NH2-terminal domain relative to the
`fingers, palm, and thumb of 31-kD COOH-
`terminal domain. The 8-kD domain is teth-
`ered to a proteolytically sensitive hinge
`region (residues 80-90) and changes from
`an open conformation in the apo structure
`to more closed conformations in the com-
`plex structures. Because of the precarious
`position of the 8-kD domain in the pol P
`apo structure, this type of conformational
`change seemed inevitable even before the
`structures of the ternary complexes were
`determined. The only other significant con-
`formational changes on complex formation
`were noticeable rigid-body movements of
`the thumb and, to a lesser degree, the
`fingers, resulting in a somewhat more tight-
`ly closed hand in the ternary com- plexes.
`A greater degree of flexibility on the part of
`the thumb subdomain has also been ob-
`served in other polymerase-DNA structures
`(12, 23, 24, 41). A least squares superpo-
`sition of the 3 1-kD domain of the apo
`structure (14) on the 31-kD domain of one
`of the ternary complex structures
`(P61)
`resulted in a root-mean-square (rms) devi-
`ation in at carbon positions of 2.5 A,
`whereas when the fingers, palm, and thumb
`subdomains were treated separately, the rms
`deviations in a carbon positions were only
`0.71, 0.69, and 0.82 A, respectively.
`The 8-kD domain has a net charge of
`(assuming neutral
`+ 10
`histidines) and
`1894
`
`binds to single stranded DNA with an
`association constant of 2 x 105 M' (26).
`It has no obvious structural equivalent in
`any of the other polymerases for which
`
`crystal structures have been determined,
`and crosslinking studies with gapped DNA
`substrates (42) suggest that the 8-kD do-
`main is probably responsible for the highly
`
`A
`
`B
`
`C
`
`Fig. 3. Stereoview ribbon diagrams (100) of (A) rat DNA pol 13, apo structure (14) and (B and C)
`ternary complexes of rat DNA pol 1 with a DNA template-primer and ddCTP in space groups P61
`and P21, respectively. In (A) the 8-kD domain is designated 8-kD, and the fingers, palm, and thumb
`subdomains of the 31-kD domain are represented by F, P, and T, respectively. A ball-and-stick
`representation highlights the ddCTP nucleotide in (B and C). In (B) the positions of the two Mg2+
`ions are marked with black spheres. These metals ions are not shown in (C) because we were
`unable to see the Mg2+ ions in electron density maps of the lower resolution P21 ternary complex
`structure.
`
`SCIENCE * VOL. 264
`
`*
`
`24 JUNE 1994
`
`

`

`ow,
`
`.................... ..........
`
`.....................
`
`woman
`
`processive short-gap filling activities found
`exclusively in pol 1B
`It has been
`(43).
`proposed that the 31-kD domain binds to
`the double-stranded segment of the tem-
`plate-primer, and the 8-kD domain binds to
`single-stranded
`the
`template
`overhang
`(44)-or in the case of binding to a short
`gap in the DNA, to the 5 '-phosphate of the
`downstream oligonucleotide (42). We see
`some evidence of this in that the 3 1-kD
`domain clearly uses its palm, fingers, and
`thumb to grasp the double-stranded seg-
`ment of the template-primer while the
`8-kD domain, although positioned quite
`differently in the two complex structures, is
`nevertheless close to where an extended
`template would be. Unfortunately, our tem-
`
`plate overhang was probably a little too short
`four bases-GAAA) to
`(only
`interact
`strongly with the 8-kD domain, causing the
`first three bases of the template to be disor-
`dered in both crystal structures. It is possible
`that, because the highly flexible 8-kD do-
`main had no template on which to anchor in
`our crystallization experiments, its position
`was determined almost entirely by crystal
`packing forces, and probably neither of the
`two conformations of the 8-kD domain seen
`in Fig. 3, B and C, is correct for template
`binding in vivo. Nevertheless, kinetic stud-
`ies of the 3 1-kD fragment alone showed that
`pol ,B can still function as a polymerase
`without the 8-kD domain, albeit at only
`about 5 percent of its normal activity (44).
`
`Table 2. Hydrogen bond interactions of 3.3 A or less between pol p and the DNA template-primer.
`
`Residue
`
`Subdomain
`
`Atom
`
`Base*
`
`Atom
`
`Distance
`
`Protein to DNA phosphate H bonds
`P-6C
`N
`P-5G
`N
`P-5G
`N
`P-5G
`N
`P-7C
`NH2
`
`02P
`02P
`01 P
`02P
`02P
`
`N
`N
`N
`
`T-9C
`T-8G
`T-8G
`
`02P
`02P
`02P
`02P
`02P
`
`2.9
`2.7
`2.9
`3.1
`2.7
`
`3.0
`3.1
`2.7
`
`2.7
`2.6
`
`Gly105
`Gly107
`Ser109
`Alal10
`Arg254
`
`Lys230
`Thr233
`Lys234
`
`Fingers
`Fingers
`Fingers
`Fingers
`Fingers
`
`Palm
`Palm
`Palm
`
`Thr292
`Tyr296
`
`Thumb
`Thumb
`
`OG1
`T-5G
`OH
`T-5C
`Protein to DNA base H bonds
`Lys234
`Palm
`T-7C
`NZ
`02
`2.9
`Tyr271
`Thumb
`OH
`P-7C
`02
`2.7
`Arg283
`NH1
`T-4G
`Thumb
`N3
`3.2
`*DNA bases are designated T or P to distinguish template bases from primer bases, respectively. Starting from
`the 5' terminus of each strand, bases are numbered 1 to 11 for the template and are numbered 1 to 7 for the
`primer. C and G represent cytosine and guanine, respectively, and atom designations follow Protein Data Bank
`nomenclature.
`
`Fingerg
`
`S
`
`Thumb
`
`FingerS
`
`Thumb
`
`8 kD
`8 kD
`Fig. 4. Omit map of the DNA template-primer and ddCTP overlayed on an a carbon diagram of the
`refined P61 pol
`structure. The view is that of Fig. 3 rotated by 900 about a horizontal axis in the
`plane of the page so that the DNA binding channel is now vertical. The template-primer sits on top
`of the palm subdomain, which is not labeled. Before all omit maps were calculated, the part of the
`structure in question was deleted from the coordinate file and the remaining partial structure was
`subjected to 200 cycles of least squares positional refinement in XPLOR (36) in order to remove
`bias from the phases.
`
`SCIENCE * VOL. 264
`
`*
`
`24 JUNE 1994
`
`In contrast to the 8-kD domain, the rest
`of the structure (the fingers, palm, and
`thumb of 3 1-kD domain, as well as the
`template-primer and ddCTP substrate) is
`virtually identical in both crystal forms of
`the ternary complex. This provides support
`for the physiological relevance of our com-
`plex crystals, at least with respect to the
`polymerase-DNA-ddCTP interactions. Also
`strengthening the argument is that, unlike
`other reported crystals of polymerase-DNA
`complexes (12, 41), our crystals were grown
`at low, near physiological salt concentra-
`tions. Finally, the fact that that a ddCMP
`was incorporated into our template-primer
`shows that the nucleotidyl transfer reaction
`did proceed, at least for one turnover, in the
`same medium from which crystals were
`eventually obtained. In that the following
`discussions do not apply to the 8-kD do-
`main of pol ,B and will be limited mostly to
`the 3 1-kD domain's interactions with
`DNA and ddCTP, we will henceforth
`refer only to the complex structure that
`has been refined to the highest resolution,
`the P61 crystal structure.
`The DNA binding channel in pol 1, just
`as in KF, RT, and RNAP, is lined with
`positively charged lysine and arginine side
`chains, and it has always been a reasonable
`assumption that their function is to stabilize
`the negatively charged backbone phos-
`phates of the DNA (45). Therefore it was
`quite surprising that, except for Arg254,
`which is hydrogen bonded to the phosphate
`of the newly incorporated ddCMP of the
`primer strand, there are no direct lysine or
`arginine interactions with the backbone
`phosphates of the DNA in our complex
`(Table 2). Instead, nearly all of the inter-
`actions of protein with DNA involve two
`different clusters of protein backbone nitro-
`gens located at the entrance to the DNA
`binding channel (Table 2). One cluster,
`consisting of four of the backbone nitrogens
`between Gly'05 and Aa1P1, is located at the
`NH2-terminal end of helix G in the fingers
`subdomain of pol 1 and interacts with the
`phosphates of the primer strand. The sec-
`ond cluster, comprising three of the five
`backbone nitrogens between Lys230 and
`Lys234, is located in a beta turn, which
`connects beta strands 3 and 4 of the palm
`subdomain and interacts with backbone
`phosphates of the template strand. The
`only other hydrogen bonded interactions
`(3.3 A or less) between pol 13 and DNA
`phosphates are between the side chains of
`Thr292 and Tyr296, located on a loop be-
`tween beta strands 6 and 7 of the thumb
`subdomain, and the backbone phosphates
`of the template strand (Table 2).
`In addition to our observations that
`there seemed to be fewer hydrogen bond
`interactions between pol 1P and DNA than
`expected (Table 2), we were also initially
`1895
`
`

`

`surprised to see that the DNA sits in the
`binding channel at a slight angle and ap-
`pears to "run into" alpha helices M and N
`of the thumb subdomain (Fig. 4). It
`is
`possible that the angle between the DNA
`axis and the apparent axis of the pol 1
`binding channel would change considerably
`if a longer template were used and, as
`proposed above, the 8-kD domain partici-
`pated in the positioning of the template-
`primer. However, the aesthetically pleasing
`observation that the base pairs of the DNA
`are parallel to the beta strands of the palm
`subdomain (Fig. 5) encourages us to believe
`that interactions of pol 13 with the double-
`stranded segment of any DNA template-
`primer will not vary much from what is seen
`in the present ternary complex structures.
`Perhaps one of the most unvarying char-
`acteristics of B-DNA is that it has a spine of
`well-ordered water molecules which inter-
`acts with the 02 of pyrimidines and the
`N3's of purines in the minor groove, and it
`has been proposed that the disruption of
`this particular water structure is the first
`step in the B-DNA to A-DNA transition
`(46, 47). In our complex structure, only
`three protein side chains come within 3.3
`A of the DNA bases, and they are all
`located in the shallow minor groove of the
`template-primer (Table 2). Two of these
`(Lys234 and Tyr27t) are hydrogen bonded to
`the 02 of a template cytidine and the 02 of
`a primer cytidine, respectively, while an-
`other (Arg283) is hydrogen bonded to the
`N3 of a template guanine. This leads us to
`propose that perhaps Lys234, Tyr271, and
`Arg283 all function to break up the water
`structure in the minor groove of the tem-
`plate-primer upon complex formation, re-
`sulting in a larger minor groove width,
`characteristic of A-DNA, at the pol 1
`active site.
`Unlike transcription factors and other
`gene regulatory DNA-binding proteins,
`polymerases must bind to DNA with little
`regard for sequence specificity. This is evi-
`dent from Table 2 where, as discussed
`above, most of the protein to DNA inter-
`actions are nonsequence-specific hydrogen
`bonds between pol 13 backbone nitrogens
`and DNA phosphate oxygens. Even what
`appear to be sequence-specific interactions
`between protein side chains and DNA