Limits of NMR Structure Determination using …brd/Teaching/Bio/asmb/current/...Limits of NMR Structure Determination using Variable Target Function Calculations: Ribonuclease T1,a

Limits of NMR Structure Determination using VariableTarget Function Calculations: Ribonuclease T1, aCase Study

Stefania Pfeiffer, Yasmin Karimi-Nejad and Heinz RuÈ terjans*

Institut fuÈ r BiophysikalischeChemie, Johann WolfgangGoethe-UniversitaÈtBiozentrum, N230, 1.OGMarie-Curie-Staûe 9D-60439, Frankfurt, Germany

Limits of NMR structure determination using multidimensional NMRspectroscopy, variable target function calculations and relaxation matrixanalysis were explored using the model protein ribonuclease T1 (RNaseT1). The enzyme consists of 104 amino acid residues and has a molecularmass of approximately 11 kDa. Primary experimental data comprise 1856assigned NOE intensities, 493 3J coupling constants and 62 values ofamid proton exchange rates. From these data, 2580 distance bounds, 168allowed ranges for torsional angles and stereospeci®c assignments for75% of b-methylene protons as well as for 80% of diastereotopic methylgroups were derived. Whenever possible, the distance restraints werere®ned in a relaxation matrix analysis including amid proton exchangedata for improvement of lower distance limits. Description of side-chainconformations were based on various models of motional averaging of 3Jcoupling constants. The ®nal structure ensemble was selected from thestarting ensemble comparing the global precision of structures withorder parameters derived from 15N relaxation time measurements.Signi®cant differences between the structure of RNase T1 in solution andin the crystal became apparent from a comparison of the two highlyresolved structures.

# 1997 Academic Press Limited

Keywords: conformational equilibria in proteins; NMR structuredetermination; ribonuclease T1; structure selection*Corresponding author

Introduction

When solving protein solution structures with theNMR methodology, the primary experimentalNMR data, i.e. NOE intensities and 3J coupling

constants, have to be translated into geometrical re-straints. While this procedure in itself is by nomeans trivial, it has been shown that the accuracyof this step critically affects the accuracy of the re-sulting structure (Olejniczak et al., 1984; Post et al.,1990; Thomas et al., 1991; Y. Liu et al., 1992; Zhao& Jardetzky, 1994). Therefore numerous attemptshave been made to improve both precision and ac-curacy of this translation step, thereby providingthe ``tightest'' restraints still compatible with theexperimental data. Among these, a relaxationmatrix analysis (Keepers & James, 1984; Boelenset al., 1988; BruÈ schweiler & Case, 1994, and refer-ences therein) so far certainly had the highest im-pact. Other methods to improve the precision ofprotein solutions structures include stereospeci®cassignments of prochiral groups (GuÈ ntert et al.,1991), and the consideration of exchange ratesupon derivation of distance information from NOEintensities (H. Liu et al., 1993). However, the trans-lation of NMR data into geometric restraints aswell as the structure calculation itself is severely

Present address: Y. Karimi-Nejad, Bijvoet Center forBiomolecular Research, Utrecht University, Padualaan 8,3584 CH Utrecht, The Netherlands.

Abbreviations used: 2D, two-dimensional; 3D, three-dimensional; COSY, correlation spectroscopy; CRMSD,coordinate root-mean-square deviation; E.COSY,exclusive correlation spectroscopy; 20GMP, 20 guanosinemonophosphate; 30GMP, 30 guanosine monophosphate;HMQC, heteronuclear multiple-quantum coherence;HSQC, heteronuclear single-quantum coherence;IRMSD, intensity root-mean-square deviation (relative);JRMSD, J coupling root-mean-square deviation; NMR,nuclear magnetic resonance; NOE, nuclear Overhausereffect; NOESY, nuclear Overhauser and exchangespectroscopy; ppm, parts per million; REDAC,redundant dihedral angle constraints; TPPI, time-proportional phase incrementation.

J. Mol. Biol. (1997) 266, 400±423

0022±2836/97/070400±24 $25.00/0/mb960784 # 1997 Academic Press Limited

JMB MS 2812 [3/2/97]

aggravated by the dynamic nature of proteins. In-ternal motions in proteins frequently occur on atime-scale not resolved by NMR spectroscopy, andNMR data therefore represent time-averaged aswell as ensemble-averaged quantities. Recently,new protocols have been introduced to account forconformational ¯exibility in the course of a proteinstructure re®nement, during which NMR restraintsare treated either as time-averaged (Torda et al.,1989, 1990, 1993; Pearlman & Kollman, 1991;Schmitz et al., 1992; Fennen et al., 1995; Nanzeret al., 1995) or as ensemble-averaged properties(Scheek et al., 1991, Kemmink et al., 1993; Bonvinet al., 1994; Bonvin & BruÈ nger, 1996). However, sofar none of the aforementioned methods can be re-garded as established, since both time-averagingand ensemble-averaging approaches may producemisleading results, such as arti®cally induced con-formational transitions (Pearlman, 1994a,b) oroverly imprecise structures (Constantine et al.,1995; Friedrichs et al., 1995). Given the problemsand artefacts associated with both approaches, weconsidered it worthwhile to explore the limits of a``standard'' structure calculation protocol with re-spect to the accurate description of conformationalheterogeneity.

The protein studied is RNase T1 (EC 3.1.27.3) fromthe fungus Aspergillus oryzae, one of the bestknown microbial enzymes (Pace et al., 1991, and re-ferences therein). Its crystal structure has been de-termined both for the free form (Martinez-Oyanedel et al., 1991) and for RNase T1 in complexwith various inhibitors, e.g. 20GMP (Arni et al.,1988) and 30GMP (Heydenreich et al., 1993; Zegerset al., 1994). The dynamic properties of the smallenzyme (11 kDa, 104 amino acid residues), havealso been studied extensively by a variety of bio-physical methods including NMR spectroscopy(Fushman et al., 1994a), ¯uorescence titrations andtemperature-jump experiments (MacKerell et al.,1991) and molecular dynamics simulations(MacKerell et al., 1988, Fushman et al., 1994b).RNase T1 contains two disulphide bridges (Cys2-Cys10 and Cys6-Cys103), upon cleavage of whichits enzymatic activity is lost. The protein catalysesRNA hydrolysis speci®cally at the 30 side of guano-sine nucleotides in a two-step mechanism. The ®rststep is a transesteri®cation to yield oligonucleo-tides with terminal guanosine 20,30-cyclic phos-phate, which in the second step is hydrolyzed toproduce terminal guanosine 30-phosphate. Theguanine binding site of the enzyme is formed byresidues 42 to 46 and 98. The speci®city for gua-nine arises from a combination of hydrogen bondsand hydrophobic interactions. The catalytically ac-tive residues are His40, Glu58 and His92, with ad-ditional support in transition state stabilisationcoming from Tyr38 and Arg77. The catalysis of theenzyme follows a general acid/base mechanism(Saenger, 1991), the details of which have been de-bated for a long time (Inagaki et al., 1981;Heinemann & Saenger, 1983; Nishikawa et al.,1987; Steyaert et al., 1990; Steyaert & Wyns, 1993;

Zegers et al., 1994). From the crystallographic stu-dies, it has been concluded that Glu58 performsthe role of the catalytic base, supported by His40engaged in a co-operative hydrogen-bondingchain. It has been proposed that His40 acts as thecatalytic base (Nishikawa et al., 1987). At the opti-mum pH of catalysis, however, His40 is proto-nated, as shown by NMR titration studies (Inagakiet al., 1981; 1985) and therefore not able to act as abase. For the catalytic acid, two proposals havebeen made: the crystallographers favour His92 asthe acid that protonates the leaving group, whilerestrained molecular dynamics simulations basedon NMR data indicated that His40 is more likelythe catalytic acid (Hoffmann et al., 1988; Schmidt,1990).

We have determined a high-resolution structure ofRNase T1, using multidimensional heteronuclearNMR spectroscopy, variable target function calcu-lations performed with the program DIANA(Braun & GoÄ , 1985; GuÈ ntert et al., 1991; GuÈ ntert &WuÈ thrich, 1991) combined with a relaxation matrixanalysis performed with the program MARDI-GRAS (Borgias & James, 1990; H. Liu et al., 1992,1993). Homo- and heteronuclear coupling con-stants have been determined for the backbone aswell as for the side-chains of the protein. Novelstrategies to generate stereospeci®c assignments ofprochiral groups as well as torsional angle re-straints from this information are presented. Thecollected data were suf®cient to generate 21 re-straints per residue. Whenever possible, internaldynamics were taken into account upon restraintgeneration, considering amide proton exchange,aromatic ring ¯ips, methyl group rotation andside-chain rotational dynamics. Direct comparisonswith experimental data have been performed atvarious stages of the calculation protocol, both forstructure selection and for structure validation. Weassessed to what extent the ®nally selected ensem-ble re¯ects the known dynamic properties ofRNase T1.

Results and Discussion

Structure calculation

Distance restraints derived from 928 resolved andstereospeci®cally assigned 2D NOE intensitieswere re®ned in ®ve consecutive cycles of relaxationmatrix calculations in conjunction with a structuregeneration starting from 50 random conformations.Finally, 796 distance ranges were obtained fromthis relaxation matrix re®nement. The 181 lowerdistance limits greater than 4 AÊ were discardedfrom the structure calculation to account for poss-ible damping of NOE intensities due to internalmobility or saturation transfer between water pro-tons and labile protons of the protein during therelaxation delay (presaturation). These two effectsare more critical for lower limits than for the upperlimits. Lower distance limits derived from overesti-

NMR Structure of Ribonuclease T1 401

JMB MS 2812 [3/2/97]

mated distances cannot be satis®ed and may dis-turb the ful®lment of other distance limits.

Including the distances derived from the 3D NOEexperiments, 1856 upper distance limits and 615lower distance limits were derived in total. Ofthese, 1464 upper and 555 lower distance limitsturned out to be relevant for the structure calcu-lation. On average, 14 upper and ®ve lower dis-tance limits per residue have been determined.

Structural information from a total of 493 vicinalcoupling constants has been included in the struc-ture calculations, accounting for an averagenumber of ®ve 3J values per residue. The exper-imentally determined 3J values comprised 793JHNHa, 57 3JHa(i)N(i � 1), 113 3JHaHb, 122 3JHbC0, 903JHbN, 14 3JC0Cg, 14 3JHaCg and 4 3JHbCd coupling con-stants. From the analysis of the side-chain coup-lings, 75% of the non-degenerated b-methyleneresonances and 80% of the resolved resonancepairs of diastereotopic methyl groups could be as-signed individually. In combination with structurecalculations, 55% of the side-chain amide protonresonances were assigned stereospeci®cally. How-ever, the stereospeci®c assignments of the side-chain amide proton resonances were taken into ac-count only after the relaxation matrix calculations.

The side-chain 3J data analysis delivered 54 w1 tor-sional angle restraints. Of these, 33 were deter-mined precisely by numerical analysis comprisingminimum and maximum ranges of �5� and�43�, respectively. The remaining 21 w1 torsionalangles were estimated qualitatively using the stag-gered rotamer model. From the 79 3JHNHa and57 3JHa(i)N(i � 1) values, allowed intervals for 66 fand 48 c torsional angles were established.Altogether, 168 torsional angle restraints were ob-tained. Detailed statistics of all structural restraintsderived from NMR data are given in Table 1. Thelocal distribution of the primary experimental datais depicted in Figure 1.

Fifty DIANA conformers were calculated applyingfour REDAC cycles, followed by a cycle of mini-mization employing experimental restraints only.In the ®nal ensemble of conformers, the confor-mation of aromatic side-chains was analyzed withrespect to the possible origin of NOE cross-peaks

arising from their ring protons using a cut-off ra-dius of 5 AÊ . Thus, a discrimination between theirsymmetry-equivalent ring protons was possible,despite their degenerate 1H and 13C chemicalshifts. The precision of the ®nal ensemble, as moni-tored by the CRMSD values, improved consider-ably upon inclusion of individually assignedHd1/Hd2 and He1/He2 NOE intensities.

To assess the information content of our input datawith respect to the correct geometry of the disul-phide bridges, we did not restrain the Cb and Sg

distances, as done frequently in addition to thecross-linking of the involved two half-cystine resi-dues (Williamson et al., 1985). Furthermore, no hy-drogen bond restraints were applied during thestructure calculations in order to avoid any bias(Havel, 1991).

Selection of conformers representing thesolution structure

The maximal pairwise CRMSD between conformerpairs as a function of the target function cut-offvalue has been used as a quantitative criterion forstructure selection (Widmer et al., 1993). A graphi-cal representation of this correlation (Figure 2) dis-plays a stepwise increase of the CRMSD valuesupon sampling of a larger conformational space by

Figure 1. The number of primary experimental dataincluding the number of measured 3J couplingconstants and the number of assigned NOE intensitiesdepending on the sequence position in RNase T1.

Table 1. Restraints used for structure calculation

Distance restraints Upper limits Lower limits

Intra-residual 319 187Sequential

(ji ÿ jj � 1) 352 140Medium-range

(2 4 ji ÿ jj4 5) 233 66Long-range 560 162Total number 1464 555

Torsional anglerestraints

f 66c 48w 54Total number 168

402 NMR Structure of Ribonuclease T1

JMB MS 2812 [3/2/97]

conformers with higher restraint violations, corre-sponding to a hierarchical clustering of conformers.An inspection of Figure 2 shows that the differencein backbone and heavy-atom CRMSD values isof the order of 0.1 to 0.3 AÊ only, suggesting thatmost side-chains are well de®ned with respect tothe backbone. It is obvious that the clusters ob-tained from both correlations are very similar. Theonly signi®cant difference was found in the regionof �2.15 to 2.35 AÊ 2 for the logarithm of target func-tion cut-off value: Here, individual clusters can bediscriminated only from an inspection of theheavy-atom CRMSD values, indicating that thedifference in the backbone conformational spacesampled by these clusters is very small. Thereforethe heavy-atom CRMSD was chosen for a furtheranalysis. The clusters I, II, III, IV and V in Figure 2include 15, 20, 34, 39 and 46 conformers, respect-ively. In order to address the question of which ofthese ensembles of DIANA conformers representsthe solution structure of RNase T1 most ade-quately, an unbiased criterion of structure selectionhas to be developed.

Simply choosing the cluster with the lowest re-straint violation might be misleading, given thefact that the violation of structural restraints ishighly dependent on the model used for the con-version of primary experimental NMR data (e.g.NOE intensities and 3J couplings) into geometricrestraints. We therefore compared our primary ex-perimental NMR data with the corresponding en-semble-averaged NMR parameters back-calculatedfrom the individual clusters, expecting that devi-

ations from the experimental data would provideus with a straightforward criterion for structure se-lection. Unfortunately, this does not hold for ourcase: the difference in the mean backbone JRMSDvalues as obtained from the ®ve different ensem-bles is of the order of 0.01 Hz for 3JHNHa as well asfor 3JHa(i)N(i � 1) coupling constants (data notshown) and therefore insigni®cant. A similar situ-ation is encountered upon back-calculation of en-semble-averaged NOE intensities: The ensemble-averaged Q-factor increases slightly, but continu-ously with an increasing number of conformers,spanning a range from 0.312 for cluster I up to0.327 for cluster V. That failure in using ensemble-averaged NMR quantities and their deviationsfrom the experimental data as a means for ensem-ble selection re¯ects the shortcomings of the em-ployed ``standard'' structure calculation protocol.Since every member of the ensemble rather thanthe ensemble as a whole has to ful®l the appliedrestraints, the accessible conformational space andthe amount of restraint violation both increaseupon inclusion of an increasing number of confor-mers, while the level of agreement with the pri-mary experimental data remains almost constant.Obviously, the difference in both precision (as re-¯ected in the maximal pairwise CRMSD) and accu-racy (as re¯ected by the restraint violations)between the individual ensembles is rather arti®-cial and not warranted by the experimental data.

If,however, the increasingconformationaldisorderassociated with the different ensembles re¯ectsgenuine internal mobility, a solution to the pro-blem should arise from a comparison of local dis-order present in the different ensembles, withrelaxation data re¯ecting dynamic properties of theprotein in a more direct way. In order to relateconformational disorder with ¯exibility, we com-pared global CRMSD values of the backbone withthe order parameters derived from a 15N relaxationtime analysis. Global CRMSD values are a sensitiveindicator for inconsistencies in long-range distancerestraints, which might be due to different mobilityof protein segments, a property accessible to relax-ation time studies. Experimentally derived S2 va-lues re¯ect the spatial restriction of rapid motionsonly and therefore represent a lower limit on theactual degree of disorder present (Friedrichs et al.,1995). The global backbone CRMSD of the structur-al clusters I to V is shown in Figure 3, togetherwith (1 ÿ S2) values originating from a 15N relax-ation time analysis of RNase T1 (Fushman et al.,1994a). Small values of (1 ÿ S2), correlated withspatially restricted motion, should coincide with alow global backbone CRMSD. From an inspectionof Figure 3, it can be seen that some of the ``hotspots'' in the protein are the fragments 48±55 and82±89, for which the relaxation data indicate an in-creased mobility. It should be noted that the(1 ÿ S2) values in fragment 82±89 show only amodest increase, as compared with the average va-lues observed for secondary structure elements.However, 15N transverse relaxation rates in this re-

Figure 2. The maximal pairwise CRMSD as a functionof the target function cut-off value for the N, Ca and C0atoms (dotted line) and all heavy-atoms (continuousline) of RNase T1. The plateaux of the graph for allheavy-atoms are denoted by roman numbers, indicatingthe corresponding structure clusters. The number ofstructures constituting the individual clusters is given inthe text. The CRMSD values were calculated after a glo-bal superposition of the polypeptide chain includingbackbone Ca, C0 and N atoms and all heavy-atoms of allresidues, respectively.


JMB MS 2812 [3/2/97]

gion exhibit pronounced exchange contributions,indicating a mobility on a time-scale not re¯ectedin the (1 ÿ S2) values (Fushman et al., 1994a). Theconformers of the ensemble I and II, however, dis-play an unrealistically small amount of confor-mational disorder in these regions. Only clusters ofhigher order (cluster III, IV and V) display a con-formational manifold that is in agreement with theexperimental 15N relaxation data. It is evident fromFigure 1 that the conformational disorder presentin the backbone of segments 48±55 and 82±89 isnot caused by a lack of experimental data. A dis-crimination between clusters III, IV and V, how-ever, could not be obtained from this type ofanalysis, which is not very surprising regardingthe fact that only the side-chain conformationalspread increases from cluster III onwards (see

above). We therefore selected the minimum num-ber of conformers that are in agreement with theavailable experimental data. The 34 DIANA con-formers in cluster III were chosen to represent thesolution structure of RNase T1. The meanglobal pairwise CRMSD value in this ensemble is0.97 AÊ for the backbone, and 1.30 AÊ for the heavyatoms, including all residues for superposition.

Quality of the solution structure

Violation of structural restraints

The target function within the ensemble of 34 con-formers ranges from 89 to 166 AÊ 2 for the individualconformers (average value: 123 � 25 AÊ 2). The aver-age violation of distance restraints is 0.08 AÊ forupper as well as for lower distance limits. Averageviolations for f, c and w1 angle restraints are 2.10�,0.25� and 2.19�, respectively. These values were ob-tained by dividing the sum of violations by thetotal number of these experimental restraintswithin the ensemble. The statistics of violated re-straints are given in Table 2. If one compares themaximal violations with the average violations ofrestraints (Table 2 and Figure 4), it is evident thatlarge violations are mostly caused by a minority ofconformations in the ensemble, the only exceptionsbeing f84 and w1

20.The torsional angle restraint for f84 allows for

positive values only, ranging from 25� to 120�.However, seven out of the 34 ®nally selected con-formers adopt negative values around ÿ80� forf84, violating the corresponding torsional angle re-straint and the upper distance limit for d(HN-Ha)by 0.36 AÊ (15%). Nevertheless, these conformerswere included in the ®nally selected ensemble, be-cause the agreement with the experimental 15Nrelaxation data and backbone 3J data improvedconsiderably for the cluster containing these con-formers. Possibly, the observed restraint violationsare caused by a conformational equilibrium. Thishypothesis is further corroborated by an exchangecontribution to the 15N linewidth of Asn84, whichis the most pronounced of all observed residues inRNase T1 (Fushman et al., 1994a).

In the case of w120, the torsional angle restraint was

de®ned quantitatively, resulting in a very tight al-lowed interval of 20�. The relatively high restraintviolation for w1

20 might be attributed either to an in-appropriate angle range from the 3J coupling con-stants caused by simpli®cations in the analysis of3J data (i.e. wrong models), or to systematic errorsin the 3J data themselves, which were not takeninto account by the error analysis. The torsionalangle restraint for w1

20 might therefore be also in-consistent with the NOE data.

The majority of signi®cant violations of distance re-straints are located in the fragment comprising re-sidues 79 to 90. It includes a b-turn and a b-bulgestructure (see Backbone conformation) and dis-plays relatively low order parameters of the N-Hvector (Figure 3). Larger violations were observed

Figure 3. (a) The global backbone CRMSD per residueversus the amino acid sequence of RNase T1 for thestructural clusters I to V in comparison with (b) valuesfor (1 ÿ S2). The order parameters S2 for the N-H vectorswere obtained from an analysis of 15N relaxation data.The error bars represent con®dence limits obtained froma Monte Carlo Approach (Fushman et al., 1994a). It isevident that the conformers constituting cluster I (black)and cluster II (red), respectively, do not represent theconformational manifold indicated by the order par-ameters. The conformational space sampled by the con-formers of the cluster III (green), however, is inagreement with the relaxation data.


JMB MS 2812 [3/2/97]

for two long-range distance restraints betweenside-chain atoms of Val33 and Trp59 and betweenTyr38 and Gly71, which are likely to be in¯uencedby slow segmental motions within the protein. Theonly strongly violated sequential distance restraintis located between the backbone atoms of Tyr42

and Asn43. The structural disorder in this regionwill be discussed in detail later.

The relatively large amount of restraint violationsre¯ected also by the target function values mightbe due to a number of factors, including themodels employed for generating structural re-straints, spin diffusion effects or multiple confor-mations. The accuracy of the structural restraints,which are sometimes very tight, is strongly depen-dent on the (in)appropriateness of the motionalmodels used for their generation. The dynamicmodels used for the interpretation of the side-chain3J couplings were not taken into account when in-terpreting the NOE data. In the case of the NOE-derived distance restraints, approximately one halfof the data resulted from the relaxation matrixanalysis considering spin diffusion and ``ensemble-averaging'', while the other half was generatedusing the isolated spin pair approximation, neglect-ing spin diffusion and internal mobility of methylgroups and aromatic rings. Given the fact thatdifferent motional models were included in theDIANA calculations, it is not surprising that insome cases inconsistencies between distancebounds and torsional angle intervals arise. How-ever, the resulting violations are rather due to gen-uine internal mobility than due to anoverinterpretation of the accuracy of experimentaldata, as becomes evident from a comparison ofFigures 3 and 4. There is a signi®cant correlationbetween restraint violations (Figure 4), confor-mational disorder as re¯ected by the globalCRMSD, and internal backbone mobility derivedfrom 15N relaxation data (Figure 3; Fushman et al.,1994a). A similar degree of correlation was foundwhen comparing S2 values back-calculated fromthe ensemble with the experimental S2 values de-rived from 15N relaxation data (data not shown).

Local precision of the structure

Angular order parameters for the f, c and w1

angles of the ensemble as a measure for the localstructural precision are shown in Figure 5. A com-parison of Figures 3 and 5 reveals that the valuesof S(f) and S(c) are not always correlated with theglobal precision and 15N relaxation data. Thisholds especially for the fragments 7±9, 10±17 and45±56: here the global CRMSD and the relaxation

Table 2. Violation of geometric restraints

Violated distance restraints: Upper Lower van der Waals

Total number (34 structures) 693 265 2540Average violation (AÊ ) 0.24 0.21 0.09Maximal violation (AÊ ) 2.05 1.65 0.63Number of consistent violations 225 102 51

Violated torsional angle restraints: f c w1

Total number (34 structures) 43 5 40Average violation (�) 7.2 6.0 3.9Maximal violation (�) 112 24 45Number of consistent violations 9 1 8

Figure 4. Deviations of (a) the back-calculated and aver-aged 3JHNHa and (b) 3JHa(i)N(i � 1) coupling constants fromthe experimentally measured values versus the aminoacid sequence of RNase T1. The 3J coupling constantswere back-calculated for every conformer and sub-sequently averaged over the ensemble of 34 conformers.For the 3JHa(i)N(i � 1) coupling constant the two differentKarplus parametrisations discussed in the text are indi-cated by a different shading. The 3JHNHa coupling con-stants were back-calculated using the same Karplusparameters as used for the restraint generation.


JMB MS 2812 [3/2/97]

data both indicate mobility, while the angularorder parameters are high. This result may be in-dicative for a correlated motion of several aminoacid residues in these regions, which does not re-quire large local torsional angle transitions. Ad-ditional evidence for this hypothesis comes fromthe fact that local motion on a time-scale compar-able with the overall rotational tumbling was de-rived from 15N relaxation data for Gly7, Ser53 andTyr56 (Fushman et al., 1994a), whereas contri-butions from faster conformational exchange pro-cesses to the relaxation rates were observed forTyr11, Asp15 and Ser17.

On the other hand, the low angular order para-meters for backbone torsional angles of Asn43 andAsn44 do not coincide with the high S2 values ofthese residues derived from 15N relaxation studies.However, a slow exchange process has to be as-sumed to explain transverse relaxation rates oftheir backbone 15N nuclei. A detailed analysis ofthe conformation of Asn43 and Asn44 is givenbelow.

Agreement with primary experimental data

From Figure 6, it is obvious that in some cases theensemble averages of the 3JHNHa and 3JNa(i)N(i � 1)

coupling constants differ considerably from theexperimentally determined couplings. The globalJRMSD values for 3JHNHa and 3JHa(i)N(i � 1) couplingconstants are 1.9 Hz and 2.3 Hz, respectively.These values signi®cantly exceed the precision ofthe experimental 3J data. The discrepancy betweenexperimental and back-calculated data re¯ects theshortcomings of the models and assumptionsunderlying the data analysis protocols at variousstages:

First of all, the methods employed to determinethe individual errors of 3J values are based on astatistical error analysis, therefore giving an esti-mate of their (noise-dependent) precision only. Sys-tematic errors originating from effects of relaxation(London, 1990; Norwood, 1993; Harbison, 1993;Norwood & Jones, 1993) have been neglected, andtherefore the accuracy of the experimental valuesin some cases might be considerably lower thanthese estimates suggest.

Second, it is not always possible to translate 3J va-lues into torsional angles unambiguously, becausethe Karplus relation is not a single-valued functionin a mathematical sense. Even with the aid of ad-ditional conformational energy criteria imposed onthe possible solutions, we frequently ended upwith multiple possibilities for backbone torsionalangles f and c. Since the DIANA algorithm doesnot allow for multiple non-overlapping torsionalangle restraints related to one torsional angle, wehad to use allowed torsional angle intervals broadenough to include all possible solutions, therebyimplicitly allowing for torsional angle values in be-tween. As a consequence, the backbone confor-mation of some residues might be predominantlydetermined by NOE distances. However, if thesedistances are inaccurate due to the effects of in-ternal motion or spin diffusion, the consequencewill be a physically unrealistic (energetically unfa-voured) virtual backbone conformation. It can beanticipated that the calculated structure will re¯ectthese possible drawbacks. Indeed, from a compari-son of Figures 4 and 6, it becomes obvious that insome cases, the ®nally calculated f angles violatethe imposed restraint insigni®cantly or not at all,without necessarily being in agreement with theunderlying experimental 3J data.

Third, when deriving backbone angle restraintsfrom the 3JHNHa and the 3JHa(i)N(i � 1) couplings, a

Figure 5. Angular order parameters for the torsionalangles (a) f, (b) c and (c) w1 as a measure of local pre-cision of the solution structure of RNase T1. The orderparameters for the f and w1 angles of proline residues,shaded black, were ®xed at ÿ75� and 19�, respectively,during structure calculations.


JMB MS 2812 [3/2/97]

rigid backbone has been assumed. This assumptionmight not always be justi®ed (Fushman et al.,1994a), and from a comparison of Figures 6 and 3,it is obvious that signi®cant deviations from exper-imental 3JHNHa values coincide with internal ¯exi-bility.

Finally, any comparison between a given proteinstructure and experimental 3J data will re¯ect the(in)accuracy of the underlying parametrisation ofthe Karplus relation. Despite the fact that thecalculated JRMSD value for the 3JHa(i)N(i � 1) coup-lings includes a smaller number of values than forthe 3JHNHa couplings (57 versus 79 residues), thecorresponding mean JRMSD value is signi®cantlyhigher for 3JHa(i)N(i � 1) coupling constants (seeabove). It seems unlikely that ¯exibility shouldexhibit differential effects on the two types of back-bone-related 3J couplings, since the backbone in-ternal motions are usually associated withcorrelated f/c ¯ips (Porter et al., 1983 and refer-ences therein; Fushman et al., 1994b). The intrinsicaccuracy of the two methods used to determine3JNNHa and 3JHa(i)N(i � 1) coupling constants is alsosimilar, since different transverse relaxation ratesof in-phase and anti-phase coherences will affectboth the J-modulated HSQC experiment and the

E.COSY-type 15N-HMQC-NOESY (Norwood, 1993;Norwood & Jones, 1993). It seems therefore justi-®ed to question the accuracy of the Karplus par-ameters reported for 3JHa(i)N(i � 1). Especially theminimal value of ÿ6.4 Hz predicted for a-helicalconformations with c � ÿ60� appears unlikely, re-garding the observation that only one of the resi-dues located in the a-helix of RNase T1 exhibits avalue 3JHa(i)N(i � 1) < ÿ2.4 Hz. In fact, a recentlypublished new parametrisation for 3JHa(i)N(i � 1)

(Wang & Bax, 1995) predicts a minimal value ofÿ1.76 Hz associated with c � ÿ60�, while theoverall shape of the new curve is very similar tothose previously reported (Pople et al., 1968). Withthe new set of Karplus parameters used for back-calculation of 3JHa(i)N(i � 1) couplings from the en-semble of DIANA conformers, the mean JRMSDvalue drops to 1.01 Hz, indicating that this set in-deed gives a better description of the torsionalangle dependence of 3JHa(i)N(i � 1) in proteins thanthe previously reported parametrisation based onINDO calculations.

In order to assess the agreement between the pri-mary NOE data and the ®nal structure, Q-factorswere computed (see Materials and Methods). Atthe end of the MARDIGRAS re®nement the ensem-

Figure 6. Average violation of geometric restraints used for the calculation of the solution structure of RNase T1

including torsional angle restraints for (a) f, (b) c and (c) w1 as well as (d) explicit upper and lower distance limits.The violations of distance restraints were calculated as averaged accumulated violations per residue relative to thenumber of used restraints, i.e. violations of a given distance restraint were averaged among the 34 conformers andthe average violations were distributed equally between the involved atoms. All these average violations were thenaccumulated for every residue and divided by the number of distance restraints per residue.


JMB MS 2812 [3/2/97]

ble-averaged Q-factor had a value of 0.31 includingall NOE intensities subjected to the relaxationmatrix re®nement. At this stage, torsional angle re-straints were not yet included in the structure cal-culation to avoid possible inconsistencies with theNOE data. Therefore, corresponding calculationswere performed for the ®nal ensemble of confor-mers, giving rise to a slightly increased ensemble-averaged Q-factor of 0.32. The ®nal ensemble-aver-aged Q-factor corresponds to a relative intensityerror of 5%, as re¯ected by the ensemble-averagedIRMSD value.

Ramachandran plot quality

The individual backbone conformations of all resi-dues in the ®nal DIANA ensemble are presented inthe Ramachandran maps shown in Figure 7. Ac-cording to an analysis with the program PRO-CHECK (Laskowski et al, 1991), 56.9% of theresidues are located in most-favoured regions,41.7% in allowed regions, and 1.4% in disallowedregions (data not shown). A comparison of thethree panels indicates that the majority of residuesfound in energetically unfavourable regions areeither glycine residues (comparing upper andmiddle panel) and/or residues exhibiting signi®-cant backbone conformational disorder, i.e. S(f)and S(c) < 0.9 (comparing upper and lower panel).All of these residues were found to be located inloop or turn regions (see Backbone conformation).

Hydrogen bonds

Hydrogen bonds were identi®ed in the DIANAconformers whenever the corresponding hydrogento oxygen distances were 43.0 AÊ and hydrogen-donor-acceptor angles 460�. In total, 299 differenthydrogen bonds were detected in the ®nal DIANAensemble. Of these hydrogen bonds, 49 were pre-sent in all members of the ensemble, and 113 hy-drogen bonds had a population 520%.

In general, the observed hydrogen bonds are in ex-cellent agreement with the experimentally ob-served amide proton exchange rates (unpublishedresults). Nearly all of the slowly exchanging amideprotons (i.e. with an exchange rate <0.004 sÿ1 atpH 5.5 and 313 K) are found to be involved in hy-drogen bonds. The only exceptions are the back-bone amide protons of residues 4, 38, 48, 62, 67, 68,85 and 102, and the side-chain amide protons ofAsn81 and Trp59. For Tyr4, the distance betweenthe amide proton and the carbonyl oxygen atom ofTyr11 slightly exceeds the limit for a hydrogenbond. All other mentioned amide protons exhibithydrogen bonds to bound water molecules (un-published results).

Disulphide bridge geometry

All DIANA conformers show the correct pattern ofthe two disulphide bridges in RNase T1, althoughthe calculations did not include any explicitly

Figure 7. Ramachandranplot for theensembleofsolutionstructures of RNase T1. Plots are shown for (a) all resi-dues, (b) all non-glycine residues and (c) all residues withangular order parameters S(f,c) > 0.9. Note that ridgesrunning parallel with the c axis at f � ÿ75� originatesfrom the four proline residues of RNase T1. Their f angleswere ®xed at ÿ75� during the calculations.


JMB MS 2812 [3/2/97]

de®ned distance ranges for the Cb and Sg atomsbetween the half-cystine residues. Obviously, theinput data contained enough information toallow for a correct formation of disulphidebonds between Cys2 and Cys10 as well as be-tween Cys6 and Cys103. However, as a conse-quence of this rather puristic approach, thedisulphide bond lengths in the ®nal ensembledeviate somewhat from the ideal value of 2.0 to2.1 AÊ : The sulphur atoms in the Cys2-Cys10bridge are 2.9 � 0.7 AÊ apart, and in the Cys6-Cys103 bridge their distance is 1.3 � 0.2 AÊ . Incase of the Cys2-Cys10 bridge, the increasedaverage disulphide bond length, as comparedwith the ideal value, is associated with a mixedchirality of this disulphide bridge (see below).Even in test calculations with explicit upper andlower distance limits for the corresponding sul-phur atoms (data not shown), ideal bondlengths were not observed.

An analysis of the side-chain torsional angles ofthe cysteine residues exhibits some remarkabledifferences in the conformations of the two disul-phide bridges (Table 3). While the Cys6-Cys103bridge uniformly adopts a right-handed confor-mation (corresponding to the ideal value ofw3 � 90�) throughout all calculated conformers,both right-handed and left-handed conformationsare found for the Cys2-Cys10 bridge. The chiralityobserved for the Cys6-Cys103 disulphide bridgewas found also in the crystal structure (Martinez-Oyanedel et al., 1991), while the Cys2-Cys10 disul-phide bridge in the crystal adopts a uniform chiral-ity with w3 � ÿ90�. However, several lines ofevidence suggest that the mixed chirality of theCys2-Cys10 disulphide bridge is a genuine prop-erty of the solution structure and not caused by alocal lack of data. For all four cysteine residues,tight w1 torsional angle restraints were obtainedfrom a quantitative analysis of w1 related 3J coup-lings. Moreover, b-methylene protons of three cy-steine side-chains were assigned stereospeci®cally,the only exception being Cys2. According to thecoupling constant analysis, Cys2 is also the only re-sidue adopting an eclipsed conformation about thew1 angle (w1 � 0 � 6�). This non-staggered rota-meric state is energetically unfavourable and mightbe regarded as an artefact of the simpli®ed mo-tional models applied for the analysis of 3J coup-lings (Karimi-Nejad et al., 1994). On the other

hand, eclipsed rotameric states were found for cy-steine residues in peptide disulphide bridges(Fishman et al., 1980; DzÏakula et al., 1996). We haveexperimental evidence that the Cys2-Cys10 disul-phide bridge experiences rotational isomerizationin solution. Reinvestigation of 15N relaxation timesof RNase T1 without presaturation of the water res-onance during NMR experiments (C. Ludwig &H. R., unpublished results) gave a T1/T2 ratio forCys2 signi®cantly larger than the average value inRNase T1, suggesting an exchange process in theneighbourhood of the N-H vector of Cys2. Theseinvestigations indicate that the N-H vector of bothCys2 and Cys10 is relatively mobile. Interestingly,the two families of conformers associated with thedifferent chiralities of the Cys2-Cys10 bridge alsoexhibit differences in the backbone conformation ofthe N-terminal b-sheet and of the ®rst part of thea-helix, which are correlated with their respectivedisulphide bridge chirality (Figure 8). From 15Nrelaxation data, it is known that this N-terminal re-gion is indeed mobile (Figure 3, lower panel), andthere is evidence from these data for a slow confor-mational exchange process located in this area ofthe protein (Fushman et al., 1994a; C. Ludwig &H. R., unpublished results.)

Description of the solution structure ofRNase T1

Backbone conformation

Figure 9 shows the backbone of the DIANA con-formers representing the solution structure ofRNase T1. The consensus secondary structure el-ements in the ensemble were identi®ed consideringthe following criteria: (1) hydrogen bonds occur-ring in more than 50% of the conformers; (2)slowly exchanging amide protons; and (3) averagebackbone torsional angles f and c. The averagevalues of the backbone torsional angles f and c ofRNase T1 are plotted in Figure 11. In order toidentify secondary structure elements from thesetorsional angles, we used idealized geometries andhydrogen-bonding pattern as given by Richardson(1981), with tolerances of �30� for backbone tor-sional angles de®ning a-helices, and a correspond-ing tolerance of �50� for b-sheets. According to theabove-mentioned criteria, the secondary structureelements observed in all crystal structures ofRNase T1 are present in solution.

Table 3. Cysteine side-chain conformation in the NMR solution structure ofRNase T1

w1(�) w2(�)a w3(�)a

Cys2 0 � 6 ÿ75 � 3/176 � 3 ÿ72 � 1/72 � 7Cys10 ÿ156 � 17 ÿ84 � 10/95 � 2Cys6 ÿ81 � 2 ÿ12 � 10 49 � 14Cys103 ÿ63 � 8 ÿ151 � 17

Average w angles of ®nal DIANA ensemble comprising 34 conformers.a Values corresponding to conformations near right-handed or left-handed disulphide

bridges with w3 � 90 or ÿ90�, respectively, were averaged individually.


JMB MS 2812 [3/2/97]

The residues Ser13 to Asp29 of RNase T1 form ana-helix with the typical hydrogen-bonding patternsincluding residues 12 to 30. The side-chains ofAsp3, Ser12 and Ser13 form partially populated hy-drogen bonds with the backbone amide protons ofresidues 13 and 14, thus ``N-capping'' the helix.The two C-terminal residues of the helix, however,exhibit backbone torsional angle values and hydro-gen-bonding patterns more characteristic for 310-

helices. The a-helix is slightly more bent at the N-terminal region than in the crystal structures ofboth complexed and uncomplexed RNase T1

(Martinez-Oyanedel et al., 1991; Arni et al., 1988;Heydenreich et al., 1993). The curvature pointsaway from the segment 65±67 towards the seg-ment 81±84.

The N-terminal segments Tyr4-Cys6 and Asn9-Tyr11 form a short antiparallel b-sheet in the crys-

Figure 8. Stereo view of the backbone of the solution structure of RNase T1 comprising residue 1 to 29 and 100 to104. The cysteine residues are labelled with their corresponding sequence number. The a-helix is drawn in red, partsof the central ®ve-stranded b-sheet in blue and the N-terminal two-stranded b-sheet in light blue. The superpositionwas carried out including the backbone atoms Ca, C0 and N. The two disulphide bridges Cys2-Cys10 and Cys6-Cys103 are shown in yellow. The Cys2-Cys10 disulphide bridge exhibits a mixed chirality connected with two dis-tinct conformations comprising residues 3 to 17 (see the text).

Figure 9. Stereo view of the complete backbone of the solution structure of RNase T1. For orientation some residuesare labelled with their corresponding sequence number. The a-helix is drawn in red, the central ®ve-stranded b-sheetin blue and the N-terminal two-stranded b-sheet in light blue. The superposition was carried out including the back-bone atoms Ca, C0 and N. Diffused disorder is obvious around residues 35, 70 and 98. Two different, clearly resolved,conformers of the backbone were observed between residues 42 and 46, 48 and 56, and 85 and 90.


JMB MS 2812 [3/2/97]

tal. In solution, these segments are less well de-®ned. The backbone conformation is more reminis-cent of two extended segments connected by abend including hydrogen bonds between residues8 and 6 as well as 7 and 5.

The central antiparallel ®ve-stranded b-sheet pre-sent in the crystal structure of RNase T1 is formedin solution. The b-sheet comprises residues Tyr38to His40, Tyr57 to Ile61, Asp76 to Asn81, Gln85 toThr91 and Phe100 to Cys103. The hydrogen-bond-ing pattern between the backbone segments 76±81and 85±91 is disturbed by insertion of residues inthe latter strand, residue 79 forming a classical b-bulge with the dipeptide 87±88. While the lo-cation of the strands II, III and V of the antipar-allel ®ve-stranded b-sheet agrees exactly withthat of the crystal structure, strand I as well asstrand IV have different locations: in the crystal,strands I and IV comprise Pro39 to Tyr42 andAsn84 to Ile90, respectively.

Between the residues 81 and 89, in solution twodistinct backbone conformational families (Figure 9)were found associated with two different values ofthe torsional angle f of Asn84, ÿ80� and 100�. Theobserved conformational heterogeneity is corrobo-rated by the results of 15N relaxation studies, ac-cording to which Asn84 is clearly involved in anexchange process (Fushman et al., 1994a).

Several b-turns could be identi®ed in the solutionstructure of RNase T1. They were classi®ed accord-ing to idealized backbone torsional angle values asgiven by Richardson (1981). The a-helix is pre-ceded by a b-turn type III, comprising residues 12to 15, and followed by a b-turn type III, formed byresidues 27 to 30. A b-turn type II0 is present in thesegment 33±36; the turn is preceded by a shortextended segment including residues 30 to 33. Ab-turn type I follows strand II of the antiparallelb-sheet, formed by residues 62 to 65. A b-turn ofthe same type is located between strands III and IVof the antiparallel b-sheet, formed by residues 81to 84. The b-turn between strand IV and V belongsto the type II family; it includes residues 92 to 95.Residues 48 to 50 form a g-turn. All of these turnswere found in the crystal structure (Martinez-Oyanedel et al., 1991) at exactly the same locations,with the exception of the turn including residues33 to 36, which is located between 34 and 37 in thecrystal structure.

Different types of conformational disorder arefound in the large loop regions of the enzyme. Theloop segments comprising residues Gly34 to Ser37,Tyr68 to Ala75 and His92 to Asn99 display a dif-fuse disorder in their backbone conformation, cor-related with a continuous distribution of theirtorsional angles f and c over a broad range. All ofthese loops are in close vicinity, located on the sur-face of the molecule towards the incoming sub-strate. For the loop composed of residues His92 toAsn99, high mobility was inferred from 15N relax-ation data. The other two loops owe their impre-cise de®nition more likely to a lack of data causedby their high glycine content. No 3J coupling was

measured for glycine residues and they cause alack of NOE data due to a low atom-packingdensity.

On the other hand, the guanine recognition loopformed by residues Lys41 to Ser54 exhibits twoclearly different conformations, related to the dis-crete disorder of the backbone of residues Tyr42 toTyr45. Signi®cant exchange contributions for the15N transverse relaxation rates were detected forresidues His40 to Asn44 (Fushman et al., 1994a),proving that the structural heterogeneity in thisloop region re¯ects a genuine conformational ex-change process going on in solution (see The activesite of RNase T1).

Figure 10 shows the mentioned differences be-tween the backbone conformation in solution andin the crystal (Martinez-Oyanedel et al., 1991). Inorder to assess the signi®cance of the structuraldifferences, it is necessary to quantify the uncer-tainty of atom positions within both the solutionstructure and the crystal structure. BecauseCRMSD values describing atomic deviations inNMR structures and B-factors describing atomicdeviations in crystal structures (B h� x2i) do notexactly match in their information content, thesemeasures of precision were normalized as given byBilleter (1992). Before normalization, the CRMSDvalues and

��Bp

values were calculated as the aver-age of the N, Ca and C0 atoms of each residue,thereby including all residues for the superpositionof structures. We ®rst assumed that differences be-tween the solution and the crystal structure thatexceed the normalized precision of each structureby more than 0.5 standard deviation can be re-garded signi®cant. Considering this criterion, sig-ni®cant differences between the two structureswere found for residues 10 to 15, 47, 63 to 66, 71and 84. For the residues 10, 47, 63 and 71, packingeffects between neighbouring protein molecules inthe crystal (Martinez-Oyanedel et al., 1991) providean explanation for the structural differences. Inseveral crystal structures of RNase T1 a chain ofwater molecules attached to residues 62 to 68 wasfound (Malin et al., 1991). However, in the calcu-lation of the solution structure, water moleculeswere not explicitly included. These water mol-ecules in the crystal might explain the structuraldifferences for residues 63 to 66. The differences ofthe backbone conformation comprising residues 10to 15 and 84 might originate from freezing singleconformations of these protein fragments in thecrystal. This hypothesis is supported by the factthat the normalized

��Bp

values of these residues(Figure 10) are much smaller than the correspond-ing normalized CRMSD values within the solutionstructure. Here, the increased CRMSD values ofthe mentioned residues re¯ect the observation ofmultiple distinct conformations rather than a lowprecision of the solution structure (see alsoFigure 9). The same situation holds for Asn43, andtherefore the precision of its coordinates in the sol-ution structure also appears rather low. As a conse-quence, the coordinate differences between the


JMB MS 2812 [3/2/97]

solution and the crystal structure for this residueseem to be statistically insigni®cant, applying theabove-mentioned criterion. On the other hand, ac-cording to the precision within each of the twoconformational families present in solution, thedifferences between solution and crystal structurewould ful®l the signi®cance criterion for one ofthese conformational families (see also Geometryof the catalytic and guanine recognition site). It isevident from this case that an overall comparisonon the basis of CRMSD values might be a some-what simpli®ed approach for the description ofstructural differences in the presence of distinctdisorder.

Side-chain conformations

Figure 11 shows the average values of the w1

angles in the solution structure of RNase T1. For 81residues of the enzyme, a torsional angle w1 is de-®ned, the remaining residues being alanine or gly-cine. Proline w1 angles were ®xed at 19� due tosteric aspects of the ring.

The majority of these 81 side-chains is well de®nedin the solution structure, as indicated by their highangular order parameters (Figure 5, bottom panel).Only 14 residues (17%) have disordered side-chains, corresponding to S(w1) < 0.9. The accuratew1 torsional angle restraints, together with thestereospeci®c assignments of the majority of b-

methylene protons, allowed for a high level ofprecision in the determination of side-chain con-formations (Figure 12 and 13).

The active site of RNase T1

Hydrogen bonds in the active site

NMR observations of bound water molecules inRNase T1 (unpublished results) revealed that someof the residues in the guanine recognition site ofthe enzyme are hydrated. In agreement with theseresults, intra-protein hydrogen bonds involvingthe corresponding functional groups were de-

Figure 10. Normalized CRMSD values within the sol-ution structure ensemble (blue) as well as between thesolution structure and the crystal structure (black), andnormalized square-roots of the B-factors of the backbone(red) are plotted versus the sequence of RNase T1. TheCRMSD values were normalized according to

CRMSDn � CRMSD ÿ hCRMSDi� C RMSDand

��Bp�n �

��Bp ÿ h ��Bp i� � ��Bp �

Figure 11. Average values of the torsional angles (a) f,(b) c and (c) w1 within the ensemble of the solutionstructures of RNase T1. The f and w1 angles of prolineresidues, shaded black, were ®xed at ÿ75� and 19�, re-spectively, during structure calculations.


JMB MS 2812 [3/2/97]

tected in only a minority of conformers of thesolution structure ensemble. These ®ndings are ingood agreement with the crystal structure ofRNase T1 (Martinez-Oyanedel et al., 1991), forwhich a hydration of the same residues has beenreported. However, contrary to the crystallo-graphic results, no hydrogen bond was found be-tween Tyr45 HN and Asn43 O or between Tyr42HZ and Asn44 Od, respectively. Furthermore, insolution the carboxylate group of Glu46 is par-tially hydrogen-bonded to Phe100 HN (28%), butnot to Asn99 Hd2. The latter hydrogen bond wasobserved in the crystal structure.

In the catalytic site, hydrogen bonds are presentbetween His40 Hd1 and Lys41 O (100%), His40 He2

and Tyr38 OZ (26%), Arg77 HZ21 and Gly74 O(44%), and Arg77 HZ22 and Asp76 O (50%). Thenumbers in parentheses denote the occurrence ofthe corresponding hydrogen bond among the 34conformers of the DIANA ensemble. Interestingly,no hydrogen bond contact was found between thecarboxylate group of Glu58 and His40 He2, both ofwhich are involved in the catalysis of the enzyme.However, the carboxylate group of Glu58 is hy-drated in solution (unpublished results), and it is

possible that a water molecule bridges the distancebetween the carboxylate group of Glu58 and His40He2.

A different hydrogen-bonding pattern was ob-served for the catalytic site in the X-ray analysis.Hydrogen bonds were found between His40 He2

and Glu58 Oe1, Arg77 He and Glu58 Oe2, Arg77HZ22 and Tyr38 OZ. The subtle differences in thehydrogen-bonding network re¯ect the differentgeometries of the active site in solution and in thecrystal, respectively (see below).

Geometry of the catalytic and the guaninerecognition site

In Table 4, average values and standard deviationsof w1 angles, calculated from the ensemble of 34DIANA conformers, are compared with the w1

angle values present in the crystal structure ofRNase T1 (Martinez-Oyanedel et al., 1991). Themost striking differences are found for the side-chains of Asn44 and Tyr45. While in the crystalstructure, Asn44 adopts a side-chain conformationoriented towards the phenolic ring of Tyr42 withw1 � ÿ 167�, a w1 value of 78� is found for Asn44 in

Figure 12. Stereo view of the guanine recognition site of RNase T1 in solution. Using all heavy-atoms for superposi-tion, (a) the ensemble of the solution structure is shown in comparison with (b) the crystal structure of nucleotide-free RNase T1 (red), of RNase T1 in complex with 20GMP (orange) and of RNase T1 in complex with 30GMP (green).Only heavy-atoms are displayed.


JMB MS 2812 [3/2/97]

solution, resulting in a different orientation of itsside-chain amide group (Figure 12). The side-chainof Try45 exhibits a w1 angle of 63� in the crystal. Insolution, Tyr45 adopts a side-chain conformationcorresponding to a w1 value of ÿ98�, resulting in acompletely different orientation of its phenolic ring(Figure 12).

A comparison of the solution structure of RNase T1

with the different crystal structures of the enzyme,corresponding to a nucleotide-free active site(Martinez-Oyanedel et al., 1991), a complex withthe nucleotide 20GMP (Arni et al., 1988) and a com-plex with the nucleotide 30GMP (Heydenreich et al.,1993), respectively, reveals some interesting fea-tures of the guanine recognition site (Figure 12). Insolution, the phenolic rings of Tyr45 and Tyr42 arestapled above each other, with their ring planes in-

cluding an angle of about 45�. Note that the dis-tance between the two rings is too far to describethis arrangement as a stacking interaction in astrict sense. The orientation of the aromatic ring ofTyr45 is similar to that found in nucleotide com-plexes of RNase T1, and it is regarded to be crucialfor guanine recognition. In the crystal structure ofthe uncomplexed RNase T1, however, the aromaticring of Tyr45 points away from its counterpart,forming a hydrogen bond to a bound water mol-ecule. So far, it has been assumed that the stapledarrangement of Tyr42 and Tyr45 occurs only in thepresence of substrate, ``sandwiching'' the guaninebase and thereby stabilizing the enzyme-substratecomplex. From the NMR structure, however, it isevident that this arrangement is already pre-formed in the absence of a substrate. Evidence for

Figure 13. Stereo view of the catalytic site of RNase T1 in solution. Using all heavy-atoms for superposition, (a) theensemble of the solution structure is shown in comparison with (b) the crystal structure of nucleotide-free RNase T1

(red), RNase T1 in complex with 20GMP (orange), and RNase T1 in complex with 30GMP (green). Only heavy-atomsare displayed.


JMB MS 2812 [3/2/97]

a rotational mobility in the tyrosine side-chains,which is necessary to allow for the substrate ac-commodation in the active site, comes from theanalysis of side-chain 3J coupling constants. Ac-cording to these data, both tyrosine side-chains dis-play restricted rotational ¯exibility about the Ca-Cb

bond, corresponding to broad Gaussian distri-butions around an average w1 angle (Karimi-Nejad,1996). From the identical 1H chemical shifts of thesymmetry-equivalent ring protons of Tyr42 as wellas Tyr45, it is also evident that both phenolic ringsundergo rapid ``ring-¯ipping''.

The conformation of the Glu46 side-chain in sol-ution is also quite interesting with respect to theorientation of its carboxylate group. Two differentconformer families associated with two differentorientations about the w2 torsional angle canbe distinguished (Figure 13). They are de®nedwith different degrees of precision. The confor-mer family with the higher precision adopts aconformation just between the two differentorientations of this side-chain found in nu-cleotide-free and complexed RNase T1 crystalstructures, whereas the conformer family withthe lower precision resembles the different orien-tations of all considered crystal structures.

A closer inspection of Figure 9 reveals that also inthe backbone of the guanine recognition loop, twofamilies of conformers can be distinguished. Thesetwo well-de®ned families again correspond todifferent backbone conformations found in crystalstructures of complexed and nucleotide-free RNaseT1, respectively. In the crystal structure ofnucleotide-free RNase T1, c43 adopts a positivevalue, while f44 is negative. Upon complexation,the values of these backbone angles change theirsign in an anti-correlated manner, leading to adifferent orientation of the Asn43-Asn44 peptidebond. This peptide bond ¯ip is necessary to allowfor hydrogen bond formation between the amideproton of Asn44 and the O-6 of the guanine basein the substrate (Martinez-Oyanedel et al., 1991). In

solution, both orientations of the Asn43-Asn44peptide bond are present, corresponding to twodifferent sets of values for c43 and f44, respect-ively. This conformational heterogeneity is in excel-lent agreement with the results of 15N relaxationstudies, and molecular dynamics simulations,during which anti-correlated transitions of the tor-sional angles c43 and f44 have been observed(Fushman et al., 1994b).

From Table 4, it is obvious that the side-chain con-formations of the catalytic site residues Tyr38,His40, Glu58, Arg77 and His92 are very similar tothose present in the crystal structure of nucleotide-free RNase T1, if one compares only w1 angle va-lues. For the conformation of the w2 torsional angle,however, the situation is different. The two histi-dine residues, His40 and His92, exhibit mean w2

values of 78 � 5� (His40) and ÿ140 � 8� (His92).Their w2 torsional angles differ signi®cantly fromthe corresponding values observed in the crystalstructure of the nucleotide-free enzyme(w2

40 � ÿ128�; w292 � ÿ58�). As a consequence, the

imidazolium ring of His40 is ¯ipped by 180� withrespect to its orientation in the crystal ofnucleotide-free RNase T1. Instead, the ring exhibitsan orientation identical with that present in crys-tals of RNase T1 complexed with 20GMP. In con-trast, the orientation of the imidazolium ring ofHis92 is ``halfway between'' those reported for thenucleotide-free enzyme and the complex with20GMP, respectively (Figure 13). The orientation ofthe phenolic ring of Tyr38 in solution differs fromits orientation as determined in all three crystalstructures (Figure 13). The w2 angle of Tyr38 in sol-ution is ÿ141 � 5�, while in all three crystal struc-tures a value around ÿ75� was found. For Glu58,conformational disorder about the w2 angle isfound in the solution structure ensemble, resultingmainly in two different orientations of its carboxy-late group. One of these orientations is similar tothat present in all three crystal structures. The ma-jority of the DIANA conformers, however, adopt asigni®cantly different w2 value, associated with theGlu58 carboxylate group pointing towards the imi-dazolium ring of His40.

As mentioned already in Introduction, controversyexists regarding the role of the residues His40,Glu58 and His92 in the catalytic mechanism ofRNase T1. The solution structure of the enzymepresented here certainly represents a high-resol-ution NMR structure, with a suf®cient amount ofboth precision and accuracy to justify detailedcomparisons with the highly resolved crystal struc-tures of the enzyme. Signi®cant structural differ-ences upon comparison with crystallographicresults were found in the recognition site as well asin the catalytic site of RNase T1. An interpretationof these ®ndings with respect to the catalytic mech-anism of the enzyme would be speculative, thereasons for this being twofold: (1) the NMR dataavailable for complexes for RNase T1 in our labora-tory are sparse compared with the extensiveamount of NMR parameters collected for the

Table 4. w1 angles of the residues in the active site ofRNase T1

Residuea w1(NMRb)(�) w1(X-rayc)(�)

Asn36 ÿ102 � 5 ÿ74Tyr38 ÿ67 � 3 ÿ76His40 69 � 2 61Tyr42 166 � 3 170Asn43 ÿ96 � 5 ÿ68Asn44 78 � 2 ÿ167Tyr45 ÿ98 � 2 63Glu46 ÿ62 � 4 ÿ66Glu58 60 � 4 55Arg77 ÿ49 � 9 ÿ59His92 ÿ77 � 5 ÿ74Asn98 ÿ179 � 12 disorderedPhe100 ÿ81 � 4 ÿ68

a Residues for which conformations in solution and in the crys-tal differ by more than 30� are given in bold letters.

b Mean w1 torsional angles and their standard deviationsamong the ensemble of 34 conformers.

c According to Martinez-Oyanedel et al. (1991).


JMB MS 2812 [3/2/97]

nucleotide-free enzyme; (2) NMR data correlatedwith pH-dependent changes in nucleotide-free aswell as in complexed RNase T1 might also beneeded. Investigations along these lines are in pro-gress.

Conclusions

A quantitative analysis of a large amount of ex-perimental NMR data enabled us to determine ahigh-resolution NMR solution structure of RNaseT1, revealing subtle differences compared withthe crystal structure of the enzyme. A remarkablefeature of the NMR structure of RNase T1 is thepresence of multiple distinct conformations inseveral well-de®ned regions of the protein, thegenuine nature of which has been assessed by acomparison with 15N relaxation data. This mightbe somewhat surprising, with respect to the factthat we have used a structure calculation protocolthat by itself does not take into account thedynamic nature of the NMR data. However,as demonstrated here, the NMR data them-selves contain enough information to resolveconformational equilibria qualitatively, evenwithout the use of rather involved time-aver-aged or ensemble-averaged methods. In orderto exploit this information, care must be takento translate the primary experimental datasuch as NOE intensities and 3J coupling con-stants into restraints that have to re¯ect boththe uncertainty of the primary experimentaldata and the dynamic nature of the proteinunder study. We have shown that a carefulanalysis of both NOE and 3J data can providea set of restraints with suitable precision. Theaccuracy of the restraints can be maintainedby incorporating models of internal motions inthe process of restraint generation. Confor-mational ¯exibility was taken into accountduring the analysis of the calculated structure.The cluster of conformers representing the sol-ution structure was selected by comparisonwith the 15N relaxation data, avoiding an un-realistic degree of precision suggested by thecluster analysis alone. The validation of thesolution structure revealed that conventionalcriteria, such as CRMSD values and restraintviolations, as well as more thorough qualitymeasures, such as stereochemical properties,are fairly well matched. Moreover, the agree-ment with the primary experimental NMRdata is also quite good, indicating a high de-gree of accuracy for the solution structure ofRNase T1.

Some caveats have to be mentioned, however, re-garding the limits of a standard NMR structuredetermination protocol. An accurate translation ofNOE data or 3J coupling constants into distancerestraints or torsional angle restraints requires ana priori knowledge of the dynamic properties ofthe molecule, thereby de®ning a circular problem.

Simpli®ed dynamic models applied indepen-dently to the NOE data and 3J coupling constantstherefore might not solve the problem of model-ling conformational equilibria completely, i.e. con-formation and population of the conformers. Theusage of these dynamic models requires a highlocal information density, e.g. in order to ®t side-chain rotameric averaging models to 3J couplingconstants. Conformational variability will there-fore in some cases remain undetected during therestraint generation. Unresolved conformationalequilibria, e.g. an overly precise structure, mayarise if a single conformation can be found thatful®l all restraints simultaneously without largeviolations.

All these factors eventually limit the accuracy ofthe solution structure of RNase T1 presented here,and this limit is re¯ected by the residual discrepan-cies between back-calculated and experimentalNMR data.

Materials and Methods

Sample preparation

Unlabelled as well as uniformly 15N and 15N/13Clabelled RNase T1 in the isoenzyme form (with a lysinein position 25) was obtained and puri®ed from recombi-nant Escherichia coli strain DH5a/pA2T1-1 as described(Quaas et al., 1988). For the labelled samples, M9 mini-mal medium was used with 15NH4Cl as nitrogen sourceand uniformly 13C-enriched glucose as the sole carbonsource.

The concentration of samples of unlabelled RNase T1

was 3.5 mM, while NMR samples of labelled RNase T1

contained 2 mM protein solutions. The solvent consistedof a mixture of 90% H2O and 10% 2H2O. In addition,15N/13C-enriched RNase T1 was dissolved in 99.996%2H2O. The pH of all samples was adjusted to 5.5 (uncor-rected for deuterium isotope effects) by small volumes ofdilute HCl and NaOH for H2O samples and 2HCl andNaO2H for 2H2O sample solutions.

NMR spectroscopy and primary experimental data

Heteronuclear NMR spectra were recorded at 313 K onBruker AMX600, DMX500 and DMX600 spectrometersequipped with triple-resonance probes. The DMX600spectrometer provided, in addition, a z-gradient facility.Quadrature detection in the indirectly detected dimen-sions was obtained using TPPI (Marion & WuÈ thrich,1983) or States-TPPI (Marion et al., 1989). HomonuclearNMR spectra were recorded on an AMX600 spec-trometer using a probe selective for 1H. Spectra wereprocessed and analysed using XWINNMR (Brunker,Rheinstetten), FELIX (Hare Research Inc., Woodinville),AURELIA (Bruker, Rheinstetten) and TRIAD (TriposInc., St. Louis) software.

NOE intensities

The following experiments were used for determi-nation of NOE intensities: 3D-1H-15N-NOESY-HSQC


JMB MS 2812 [3/2/97]

(Stonehouse et al., 1994; Kay et al., 1992); 3D-1H-13C-NOESY-HMQC (Ikura et al., 1990) selective for aromaticcarbon atoms; 3D-1H-13C-HSQC-NOESY (Zuiderweget al., 1990; Vuister & Bax, 1992) selective for aliphaticcarbons atoms with mixing times of 100 ms, 150 ms and120 ms, respectively. In order to achieve adequate resol-ution, the 3D NOE exeriment for aliphatic carbon atomswas folded twofold in the carbon dimension using theStates-TPPI method for sign discrimination. All 3D ex-periments were recorded with 64 increments in the 13Cand 15N dimension, respectively. The data were sub-jected to linear prediction and zero-®lled up to a resol-ution of approximately 15 Hz/point and 23 Hz/pointfor the indirect 15N and 13C dimensions, 11 Hz/point inindirect 1H dimensions, and 6 Hz/point for the direct1H dimension.

In addition, a series of 2D-1H-1H-NOESY spectra wasacquired with mixing times of 60, 90, 120 and 150 ms(Jeener et al., 1979; Otting et al., 1986; Marion & Bax,1988). For each of the 2D NOESY spectra, 800 incre-ments in t1 were recorded, sampling 4096 real datapoints in t2. After zero-®lling and Fourier transform-ation, a resolution of 3.7 Hz/point in both dimensionswas achieved.

3J coupling constants

The 3JHNHa coupling constants were obtained from J-modulated 2D-1H-15H-COSY experiments (Billeter et al.,1992). A series of eight 1H-15N-COSY spectra, correspond-ing to eight different delays t for the evolution of 3JHNHacouplings, was recorded. The delays were varied fromtmin � 11.181 ms to tmax � 137.181 ms, using a stepsize of18 ms; 160 and 2048 real data points were recorded in t1

and t2, respectively. The parameters of a model functionwere then numerically ®tted to the decay of the auto-peakintensities as a function of t (Billeter et al., 1992), one ofthe parameters being the 3JHNHa coupling constant.Individual estimates for the errors in the 3J valueswere determined by multivariate error analysis(Clifford, 1973).

The 3JHa(i)N(i � 1) and 3JNHb coupling constants were ex-tracted from a 3D-1H-15N-HMQC-NOESY experiment(Wider et al., 1989). The 3JHaHb,

3JC0Hb and 3JC0Cg couplingconstants were determined from three different E.COSY-type experiments: 3D-Soft-HCCH-E.COSY (Eggenbergeret al., 1992a), 3D-Soft-HCCH-COSY (Eggenberger et al.,1992a) and 3D-Soft-HCCC-COSY (Schwalbe et al., 1993).The 3JHaCg coupling constants in valine side-chains wereobtained from a 13C-®ltered 2D-1H-1H-NOESY with a se-lective decoupling of the 13Ca and 13C0 resonances. Ex-perimental details and acquisition parameters of all theseexperiments have been reported (Karimi-Nejad et al.,1994). A 3D-1-H-13C-CT-HSQC-NOESY experiment withselective decoupling of Cb and C0 resonances in o3 wasused for measuring 3JHbCd coupling constants of leucineresidues (Eggenberger et al., 1992b).

The E.COSY-type spectra were evaluated using the pre-viously described J-convolution method, thereby ob-taining precise 3J values, their individual experimentalerrors and R-factor values as a measure of ®t quality(Karimi-Nejad et al., 1994). In cases where overlap orinsuf®cient signal to noise precluded the application ofthe J-convolution method, estimates of the 3J valueswere obtained from the relative displacement of thetraces corresponding to the upper and lower half ofthe E.COSY multiplet.

Amide proton exchange rates

Slow amide proton exchange was monitored from aseries of 50 2D-1H-15-HSQC spectra (Bodenhausen &Ruben, 1980) after addition of 2H2O, while the fast ex-change with the solvent was observed in a series ofMEXICO spectra (Gemmecker et al., 1993) with mixingtimes of 25, 50, 75, 100, 150 and 200 ms recorded in H2O.The rate constants were obtained by non-linear, least-squares ®tting of the exchange rate constants to the time-dependent intensities, using the Levenberg-Marquardtalgorithm.

Geometric restraints for the structure calculation

Initial distance restraints

The assignment of the 3D NOE intensities was per-formed iteratively using the facilities of the TRIPOSsoftware (Tripos Inc., St. Louis) and the complete as-signment of the 1H, 15N and 13C resonances of RNaseT1 (Pfeiffer et al., 1996). For the generation of an initialstructure, we used an isotropic motional model, andthe isolated spin pair approximation for the interpret-ation of NOE intensities. Three different calibration dis-tances were used, corresponding to the three different3D NOE experiments mentioned above: distance limitsderived from the 3D-1H-15N-NOESY-HSQC spectrumand from the 3D-1H-13C-HSQC-NOESY spectrum (selec-tive for aliphatic carbons) were calibrated with charac-teristic distances found in secondary structure elementsof proteins (a-helix: d(HN

i -HNi ÿ 1) � 2.8 AÊ ; b-sheet:

d(Hai -Haj) � 2.3 AÊ ). Distance restraints obtained from the

3D NOE experiment selective for aromatic carbon atomswere calibrated using a ®xed distance in the singletryptophan residue (d(HZ2-Hz2) � 2.5 AÊ ). Distances invol-ving methyl protons or symmetry-equivalent ring pro-tons were subjected to appropriate pseudoatomcorrections. Subsequently, distance ranges were gener-ated by addition and subtraction of an error of 10% ofthe derived distances. Using these distance ranges, wegenerated an initial ensemble of conformers. Severalcycles of assignment and structure calculation followed,during which structure ensembles of low target functionwere used as a reference to reduce ambiguities in theNOE assignments. Finally, the ®lter distance for assign-ing NOE cross-peak intensities was decreased from7.5 AÊ to 5.5 AÊ .

Torsional angle restraints and stereospecific assignments

f and c torsional angles were calculated from 3JHNHaand 3JHa(i)N(i � 1) coupling constants using the followingKarplus parametrisations:

3JHNHa�f� � 6:7 co s2�fÿ 60� ÿ 1:3�fÿ 60� � 1:5 �1�

3JHa�i�N�i�1��c� � ÿ 5:1 co s2�cÿ 120�� 2:2 co s�cÿ 120� � 0:9 �2�

(Ludvigsen et al., 1991; Pople et al., 1968). Due to theambiguity of the Karplus relation, up to four valuesof torsional angles will be obtained from each ofthese equations, assuming that the protein backboneadopts a single conformational state. The confor-mational energy of the resulting backbone confor-mations was estimated from an adiabatic f,c energymap. Only energetically favoured f/c combinationswere considered to de®ne allowed ranges for the


JMB MS 2812 [3/2/97]

backbone torsional angles. Details of the method havebeen described (Karimi-Nejad, 1996).

In cases where low-energy backbone conformations withpositive f angles resulted from the above described pro-cedure, the intra-residual d(HN-Ha) was used as an ad-ditional criterion (Ludvigsen & Poulsen, 1992). Apositive f angle was assumed when this distance wassigni®cantly shorter than the average d(HN-Ha) in the a-helix of the protein, i.e. d(HN-Ha) < 2.5 AÊ . Wheneverd(HN-Ha) was not available from the NOE measure-ments, positive f angles were taken into account.

Since the program DIANA (Version 2.8) does not handlemultiple non-overlapping intervals as a restraint for onetorsional angle, we used ranges including all possibleconformations, including an error margin of �10�.

No torsional angle restraint was de®ned for residues forwhich only the 3JHa(i)N(i � 1) coupling was available. Forthose residues for which only 3JHNHa coupling constantswere determined, f torsional angle restraints were de-®ned in a qualitative manner, using ÿ80� 4 f4 ÿ 40�for 3JHNHa < 5 Hz andÿ160� 4 f4 100� for 3JHNHa > 8 Hz(Case et al., 1995).w1 torsional angle restraints were obtained by ®tting

models for the side-chain conformation to the exper-imental 3J data, using a strategy analogous to the pre-viously described conformational analysis of valine side-chains in RNase T1 using three different models for thedescription of side-chain rotational dynamics (Karimi-Nejad et al., 1994). The following Karplus parametrisa-tions were used during the numerical analysis:

3JHaHb2�w1� �9:5 co s2�w1 ÿ 120�ÿ 1:6 co s�w1 ÿ 120� � 1:8 �3�

3JHaHb3�w1� � 9:5 co s2�w1� ÿ 1:6 co s�w1� � 1:8 �4�

3JC 0Hb2�w1� � 7:2 co s2�w1� ÿ 2:04 co s�w1� � 0:6 �5�

3JC 0Hb3�w1� �7:2 co s2�w1 � 120�ÿ 2:04 co s�w1 � 120� � 0:6 �6�

3JNHb2�w1� � ÿ 4:4 co s2�w1 � 120�� 1:2 co s�w1 � 120� � 0:1 �7�

3JNHb3�w1� � ÿ 4:4 co s2�w1 ÿ 120�� 1:2 co s�w1 ÿ 120� � 0:1 �8�

(DeMarco et al., 1978a, b; Fishman et al., 1980). For everyresidue the most appropriate model was chosen from anevaluation of ®t quality determined from ÿ-function va-lues (Press et al., 1989).

Stereospeci®c assignments of the 1H chemical shifts of b-methylene protons were carried out by performing theabove-mentioned ®t procedures for both possible assign-ments, selecting the assignment that gave the lower errorfunction. However, only those residues were stereospeci-®cally assigned for which the assignment turned out tobe model-independent.

The resulting side-chain conformations were translatedinto allowed w1 angle ranges using the followingstrategy: (1) for side-chains adopting a single rotamer,an error margin of �10� was added to the torsionalangle obtained from the ®t procedure. This error

range seems to be warranted regarding possible inac-curacies of Karplus parameters as well as systematicexperimental errors in the 3J values (Karimi-Nejadet al., 1994). (2) Whenever Gaussian distributionsabout a single w1 angle explained experimental databest, the allowed interval of w1 was adjusted accord-ing to the width of Gaussian distribution. (2) Resi-dues, for which the 3J data indicated equilibriabetween rotameric states, were not restrained in w1

during the structure calculations.In our opinion, this procedure is a more accurate way of

dealing with local mobility than direct J-re®nementmethods (Garrett et al., 1994), during which every gener-ated conformer is forced to agree with the experimental Jvalue, thereby disregarding its nature as a time-averagedor ensemble-averaged quantity. Furthermore, our ap-proach circumvents the already mentioned dif®cultiesknown to be associated with time-averaged or ensemble-averaged protocols (Nanzer et al., 1995; Pearlman, 1994a;Constantine et al., 1995).

Side-chains for which all couplings were only estimates,obtained from the displacement within an E.COSY multi-plet, were not subjected to the above-mentioned pro-cedures. Instead, restraints for their w1 angles werede®ned around the most probable staggered rotamer,using a tolerance of �30�.

Stereospeci®c assignments of the valine methyl groupsin RNase T1 have been reported (Karimi-Nejad et al.,1994). The resonances of diastereotopic methyl groupsin leucine side-chains were stereospeci®cally assignedby inspection of their 3JHbCd couplings. In the case ofLeu62, it was possible to determine a w2 torsionalangle restraint by identi®cation of the w2 staggeredrotamer.

Structure calculation and distance refinement

Structures were generated using the program DIANA(GuÈ ntert et al., 1991) including the REDAC strategy(GuÈ ntert & WuÈ thrich, 1991) and the program GLOMSAfor additional structure aided stereospeci®c assignments(GuÈ ntert et al., 1991). Default weighting factors were at-tributed to the different restraint categories. The twoknown disulphide bridges between positions 2 and 10 aswell as in positions 6 and 103 were included in the top-ology of the molecule. All peptide bonds were ®xedtrans, except for Pro39 and Pro55, for which cis peptidebonds were identi®ed from an inspection of theirsequential NOE cross-peak intensities (WuÈ thrich, 1986)and their nitrogen chemical shifts (Schmidt et al., 1991).All histidine, arginine and lysine residues were regardedto be positively charged, while the glutamate and as-partate side-chains were treated as negatively chargedresidues, in agreement with previously reported pH-de-pendent 1H-NMR investigations (RuÈ terjans et al., 1969;RuÈ terjans & Pongs, 1971; Arata et al., 1976, 1979;Menke, 1984).

The structure ensemble calculated from distances de-rived from 3D NOE data was utilized to assign the 2DNOE cross-peaks in the homonuclear spectra, assumingthat the maximal distance giving rise to an NOE is5.5 AÊ . With the exception of non-stereospeci®cally as-signed protons and methyl groups (with resolvedchemical shifts), all 2D NOE intensities with a singleassignment were included in a relaxation matrix calcu-lation using the program MARDIGRAS (Borgia &James, 1990). Unresolved diastereotopic methylene pro-tons and methyl groups were treated as pseudoatoms.


JMB MS 2812 [3/2/97]

NOE intensities with ambiguous assignments were ex-cluded from these calculations, because the contributionof each possible NOE interaction to be observed NOEintensity cannot be estimated. During the relaxationmatrix calculations internal ¯exibility was taken intoaccount for methyl groups and aromatic rings (H. Liu,et al., 1992). The 18-site jump model of MARDIGRAS(free rotational diffusion) was used both for intra- andinter-residual methyl distances instead of a three-sitejump model, since we observed that the algorithm ofDIANA does not necessarily arrange the methyl pro-tons in energetically favourable staggered positions.Since the inertia tensor calculated from the crystalstructure (Martinez-Oyanedel et al., 1991) is 1:1.14:1.27,we assumed isotropic rotational tumbling of the mol-ecule with a correlation time of 5.4 ns (Fushman et al.,1994a). The exchange of amide protons with the sol-vent was taken into account to obtain more accuratelower distance limits from the relaxation matrix analy-sis (Liu et al., 1993). A total of 62 values for exchangerate constants of 15N bound protons of RNase T1 wasincluded in the relaxation matrix calculation. Exper-imental and theoretical cross-relaxation rates werescaled to each other using experimental NOE intensitiesrelated to covalently ®xed distances.

The re®nement of distance restraints using MARDI-GRAS comprised several cycles of matrix calculationin conjunction with structure calculations usingDIANA. The 2D NOE experiment with tm � 150 mswas used for the relaxation matrix analysis because itprovided the largest amount of NOE cross-peaks andthe best signal to noise ratio. The 2D NOE exper-iments with shorter mixing times were used only tocon®rm stereospeci®c assignments obtained from 3Jcouplings and for an estimation of a few distance re-straints involving very fast exchanging 15N-boundside-chain protons of Arg77 and the hydroxyl protonof Tyr11. In the case of overlap of 2D NOE intensi-ties, the upper distance limits derived from the corre-sponding resolved 3D NOE intensities were usedduring the structure calculations. The lower distancelimits were excluded in this procedure to avoid dis-tortions in the re®nement of 2D NOE-derived dis-tances. In order to account for the conformationalspread among the DIANA ensembles, the relaxationmatrix re®nement was repeated using ten individualconformers of low target function values on the aver-age. Error bounds for the distances were determinedby averaging the calculated distances over the differ-ent model structures, subsequently using the standarddeviation for computation of distance ranges. Upperdistance limits were taken from the relaxation matrixre®nement excluding the matrix of chemical exchange,while lower distance limits were taken from relax-ation matrix calculations including amide proton ex-change rates (Liu et al., 1993). It should beemphasized that this procedure is not an ensemble-averaging in the thermodynamic sense. Only themaximal distance error resulting from the use ofdifferent model structures was taken into account.

Since the assumptions about internal dynamics in-herent to the MARDIGRAS calculations neglect ro-tational dynamics about the Ca-Cb bond in the caseof resolved methylene resonances, possible errors maybe introduced for distances to b-methylene protonsinvolved in torsional oscillations. In the analysis of 3Jdata, however, this type of ¯exibility was explicitlytaken into account, and the w1 angle restraints de-rived from this analysis therefore re¯ect this assump-

tion. In order to maintain internal consistency withineach category of restraints, we have therefore ex-cluded the 3J-derived torsional angle restraints fromthe DIANA calculations during the distance re®ne-ment using MARDIGRAS.

The ®nal ensemble of structures was further subjected toa restrained energy minimization using the GROMOSforce ®eld (van Gunsteren & Berendsen, 1987). An analy-sis of the energy-minimized structures reveals no signi®-cant changes on the coordinates. This is not surprisingregarding the huge number of restraints (PDB entryR1YGWMR) allowing for only small adaptations of thegeometry (Billeter et al., 1990; Berndt et al., 1996).

Analysis of structures

Structures (PDB entry 1YGW) were visualized on SiliconGraphics Indy workstations using the program SYBYL(Tripos Inc, St. Louis). CRMSD values were obtainedusing facilities and de®nitions of the program DIANA(GuÈ ntert et al., 1991), superimposing all residues to de-scribe global precision, whereas local precision was de-scribed by CRMSD values resulting from a superpositionof tripeptide segments. Backbone CRMSD values in-cluded N, Ca and C0 atoms.

The mean RMSD values for coupling constants (JRMSD)were computed from the deviation of the ensembleaverages of the back-calculated 3J coupling constantsfrom the observed 3J value:

JRMSD ��1

N

XN

j�1

�Jexpj ÿ hJbackj i�2vuut

where N is the number of included residues or couplingconstants, respectively. Local absolute deviations of en-semble-averaged values from experimental 3J data werecalculated, giving a measure of local structure accuracy.

Angular order parameters S(y) (Hyberts et al., 1992) wereused to assess the conformational spread among the Ncalculated structures.

Back-calculation of 2D NOE intensities was carried outusing the CORMA algorithm (Borgias & James, 1990),employing the same parameters and assumptions asused during the MARDIGRAS calculations. The Q-fac-tors (Withka et al., 1992) and the IRMSD values be-tween back-calculated and experimentally determinedNOE intensities of a single model structure were calcu-lated for all NOE intensities included in the re®nement.For computation of ensemble-averaged Q-factors andIRMSD values, respectively, the theoretical NOE inten-sity was replaced by its ensemble average. This aver-aging over relaxation rates is different to averagingover distances in the ensemble of structures (Schmitzet al., 1992).

Acknowledgements

We thank the Deutsche Forschungsgemeinschaft for agrant (RU-145/8-7). We are grateful to Harald ThuÈ ring,Stefan Geschwindner, Norman Spitzner, Michael War-was, Matthias Wirth and Azita Mesgarzadeh for prepar-ing samples of RNase T1. We further thank D. Fushmanfor enlightening discussions about relaxation time analy-sis. We are indebted to John Markley for a pre-print ofDzÏakula et al., (1996). We thank the referees for various


JMB MS 2812 [3/2/97]

valuable hints. S. P. thanks Peter GuÈ untert and He Liufor helpful discussions concerning the possibilities of theprograms DIANA and MARDIGRAS, respectively. Y.K.-N acknowledges a grant from the GraduiertenfoÈrder-ung NRW, Germany. This work forms part of the PhDthesis of S. P. and Y. K.-N.

References

Arata, Y., Kimura, S., Matsuo, H. & Narita, K. (1976).Proton magnetic resonance studies of ribonucleaseT1. Assignment of His-40 peak and analysis ofactive site. Biochem. Biophys. Res. Commun. 73, 133±140.

Arata, Y., Kimura, S., Matsuo, H. & Narita, K. (1979).Proton and phosphorus nuclear magnetic resonancestudies of ribonuclease T1. Biochemistry, 18, 18±24.

Arni, R., Heinemann, U., Tokuoda, R. & Saenger, W.(1988). Three-dimensional structure of the ribonu-clease T1-2

0GMP complex at 1.9 AÊ resolution. J. Biol.Chem. 263, 15358±15368.

Berndt, K. D., GuÈ ntert, P. & WuÈ thrich, K. (1996). Confor-mational sampling by NMR solution structures cal-culated with the program DIANA evaluated bycomparison with long-time molecular dynamics cal-culations in explicit water. Proteins: Struct. Funct.Genet. 24, 304±313.

Billeter, M. (1992). Comparison of protein structuresdetermined by NMR in solution and by X-ray dif-fraction in single crystals. Quart. Rev. Biophys. 25,325±377.

Billeter, M., Schaumann, Th., Braun, W. & WuÈ thrich, K.(1990). Restrained energy re®nement with twodifferent algorithms and force ®elds of the structureof the a-amylase inhibitor tendamistat determinedby NMR in solution. Biopolymers, 29, 695±706.

Billeter, M., Neri, D., Otting, G., Qian, Y. Q. &WuÈ thrich, K. (1992). Precise vicinal coupling con-stants 3JHNa from nonlinear ®ts of J-modulated [15N,1H]-COSY-experiments. J. Biomol. NMR, 2, 257±274.

Bodenhausen, G. & Ruben, D. J. (1980). Natural abun-dance nitrogen-15 NMR by enhanced heteronuclearspectroscopy. Chem. Phys. Letters, 69, 185±189.

Boelens, R., Koning, T. M. G. & Kaptein, R. (1988). De-termination of biomolecular structures from proton-proton NOE's using a relaxation matrix approach.J. Mol. Struct. 173, 299±311.

Bonvin, A. M. J. J. & BruÈ nger, A. (1996). Do NOE dis-tances contain enough information to assess therelative populations of multi-conformer structures?.J. Biomol. NMR, 7, 72±76.

Bonvin, A. M. J. J., Boelens, R. & Kaptein, R.(1994). Time- and ensemble-averaged direct NOErestraints. J. Biomol. NMR, 4, 143±149.

Borgias, B. A. & James, T. L. (1990). MARDIGRAS±Aprocedure for matrix analysis of relaxation for dis-cerning geometry of an aqueous structure. J. Magn.Reson. 87, 475±487.

Braun, W. & GoÄ , N. (1985). Calculation of protein con-formations by proton-proton distance restraints. Anew ef®cient algorithm. J. Mol. Biol. 186, 611±626.

BruÈ schweiler, R. & Case, D. A. (1994). Characterisationof biomolecular structure and dynamics by NMRcross relaxation. Prog. NMR Spectrosc. 26, 27±58.

Case, D. A., Dyson, H. J. & Wright, P. E. (1995). Use ofchemical shifts and coupling constants in nuclearmagnetic resonance structural studies on peptidesand proteins. Methods Enzymol, 239, 392±416.

Clifford, A. A. (1973). A Handbook of Multivariate ErrorAnalysis, pp. 71 ff, Applied Science Publishers Ltd,London.

Constantine, K. L., Mueller, L., Andersen, N. H., Tong,H., Wandler, C. F., Friedrichs, M. S. & Bruccoleri,R. E. (1995). Structural and dynamic properties of ab-hairpin forming linear peptide. 1. Modeling usingensemble-averaged constraints. J. Am. Chem. Soc.117, 10841±10854.

DeMarco, A., LlinaÂs, M. & WuÈ thrich, K. (1978a). Analy-sis of 1H-NMR spectra of ferrichrome peptides. I.The non-amide protons. Biopolymers, 17, 617±636.

DeMarco, A., LlinaÂs, M. & WuÈ thrich, K. (1978b). 1H-15Nspin-spin couplings in alumichrome. Biopolymers,17, 2727±2742.

DzÏakula, Z., Westler, W. M. & Markley, J. L. (1996).Continuous probability distribution (CUPID) analy-sis of potentials for internal rotations. J. MagnReson. ser. B, 111, 109±126.

Eggenberger, U., Karimi-Nejad, Y., ThuÈ ring, H.,RuÈ terjans, H. & Griesinger, C. (1992a). Determi-nation of Ha,Hb and Hb,C0 coupling constants in13C-labeled proteins. J. Biomol. NMR, 2, 583±590.

Eggenberger, U., Schmidt, P., Sattler, M., Glaser, S. J. &Griesinger, G. (1992b). Frequency-selective decou-pling with recursively expanded soft pulses in mul-tinuclear NMR. J. Magn. Reson. 100, 604±610.

Fennen, J., Torda, A. E. & van Gunsteren, W. F. (1995).Structure re®nement with molecular dynamics andBoltzmann-weighted ensembles. J. Biolmol. NMR, 6,163±170.

Fishman, A. J., Live, D. H., Wyssbrod, H. R., Agosta,W. C. & Cowburn, D. (1980). Torsion angles incysteine bridges of oxytocin in aqueous solution.Measurement of circumjacent vicinal coupling con-stans between 1H, 13C and 15N. J. Am. Chem. Soc.102, 2533±2539.

Friedrichs, M. S., Stouch, T. R., Bruccoleri, R. E.,Mueller, L. & Constantine, K. L. (1995). Structuraland dynamic properties of a b-hairpin-forming lin-ear peptide 2. 13C NMR relaxation analysis. J. Am.Chem. Soc. 117, 10855±10864.

Fushman, D., Weisemann, R., ThuÈ ring, H. & RuÈ terjans,H. (1994a). Backbone dynamics of ribonuclease T1

and its complex with 20GMP studied by two-dimen-sional heteronuclear NMR spectroscopy. J. Biomol.NMR, 4, 61±78.

Fushman, D., OhlenschlaÈger, O. & RuÈ terjans, H. (1994b).Determination of backbone mobility of ribonucleaseT1 and its 20GMP complex using moleculardynamics simulations and NMR relaxation data.J. Biol. Struct. Dynam, 11, 1377±1402.

Garrett, D. S., Kuszewski, J., Hancock, T. J., Lodi, P. J.,Vuister, G. W., Gronenborn, A. & Clore, M. (1994).The impact of direct re®nement against three-bondHN-CaH coupling constants on protein structuredetermination by NMR. J. Magn. Reson ser. B, 104,99±103.

Gemmecker, G., Jahnke, W. & Kessler, H. (1993).Measurement of fast proton exchange rates in isoto-pically labeled compounds. J. Am. Chem. Soc. 115,11620±11621.

GuÈ ntert, P. & WuÈ thrich, K. (1991). Improved ef®ciencyof protein structure calculation from NMR datausing the program DIANA with redundant dihe-dral angle constraints. J. Biomol. NMR, 1, 447±456.

GuÈ ntert, P., Braun, W. & WuÈ thrich, K. (1991). Ef®cientcomputation of three-dimensional protein structuresin solution from NMR data using the program


JMB MS 2812 [3/2/97]

DIANA and supporting programs CALIBA,HABAS and GLOMSA. J. Mol. Biol. 217, 517±530.

Harbison, G. S. (1993). Interference between J-couplingsand cross-relaxation in solution NMR spectroscopy:consequences for macromolecular structure de-termination. J. Am. Chem. Soc. 115, 3026±3027.

Havel, T. F. (1991). An evaluation of computationalstrategies for use in the determination of proteinstructure from distance constraints obtained bynuclear magnetic resonance. Prog. Biophys. Mol. Biol.56, 43±78.

Heinemann, U. & Saenger, W. (1983). Crystallographicstudy of mechanism of ribonuclease T1 catalyzedspeci®c RNA hydrolysis. J. Biomol. Struct. Dynam. 1,523±538.

Heydenreich, A., Koellner, G., Choe, H.-W., Cordes, F.,Kisker, C., Schindelin, H., Adamiak, R., Hahan, U.& Saenger, W. (1993). The complex between ribo-

nuclease T1 and 30GMP suggests geometry of enzy-matic reaction path. An X-ray study. Eur. J. Biochem.218, 1005±1012.

Hoffmann, E., Schmidt, J., Simon, J. & RuÈ terjans, H.(1988). Solution structure of RNase T1 and its com-plexes with nucleotides. Nucleosides Nucleotides, 7,757±761.

Hyberts, S. G., Goldberg, M. S., Havel, T. F. & Wagner,G. (1992). The solution stucture of eglin c based onmeasurements of many NOE's and coupling con-stants and its comparison with X-ray structures.Protein Sci. 1, 736±751.

Ikura, M., Kay, L. E., Tschudin, R. & Bax, A. (1990).Three-dimensional NOESY-HMQC spectroscopy ofa 13C-labelled protein. J. Magn. Reson. 86, 204±209.

Inagaki, F., Kawano, Y., Shimada, I., Takahashi, K. &Miyazawa, T. (1981). NMR study of microinviron-ments of histidine residues of ribonculease T1 andcarboxymethylated ribonuclease T1. J. Biochem.(Tokyo), 89, 1185±1195.

Inagaki, F., Shimada, I. & Miyazawa, T. (1985). Bindingmodes of inhibitors to ribonuclease T1 studied bynuclear magnetic resonance. Biochemistry, 24, 1013±1020.

Jeener, J., Meier, B. H., Bachmann, P. & Ernst, R. R.(1979). Investigation of exchange processes by two-dimensional NMR spectrsocopy. J. Chem. Phys. 71,4546±4553.

Karimi-Nejad, Y. (1996). NMR-spektroskopische Unter-suchungen zur Struktur der Ribonuclease T1 undihrer Komplexe mit 20 und 30 GMP. PhD thesis,University of Cologne, Germany.

Karimi-Nejad, Y., Schmidt, J. M., RuÈ terjans, H.,Schwalbe, H. & Griesinger, G. (1994). Conformationof valine side chains in ribonuclease T1 determinedby NMR studies of homonuclear and heteronuclear3J coupling constants. Biochemistry, 33, 5481±5492.

Kay, L. E., Keifer, P. & Saarinen, T. (1992). Pure absorp-tion gradient enhanced heteronuclear single quan-tum correlation spectroscopy with improvedsensitivity. J. Am . Chem. Soc. 114, 10663±10665.

Keepers, J. W. & James, T. (1984). A theoretical study ofdistance determination from NMR. Two-dimen-sional nuclear Overhauser effect spectra. J. Magn.Reson. 57, 404±426.

Kemmink, J., Van Mierlo, C. P. M., Scheek, R. M. &Creighton, T. E. (1993). Local structure due to anaromatic-amide interaction observed by protonnuclear magnetic resonance spectroscopy in pep-tides related to the N terminus of bovine pancreatictrypsin inhibitor. J. Mol. Biol. 230, 312±322.

Laskowski, R. A., MacArthur, M. W., Moass, D. S. &Thornton, J. (1991). PROCHECK, a program tocheck the stereochemical quality of proteinstructures. J. Appl. Crystalog. 24, 946±950.

Liu, H., Thomas, P. D. & James, T. L. (1992). Averagingof cross-relaxation rates and distances for methyl,methylene and aromatic ring protons due to motionor overlap: extraction of accurate distances itera-tively via relaxation matrix analysis of 2D NOEspectra. J. Magn. Reson. 98, 163±175.

Liu, H., Kumar, A., Weisz, K., Schmitz, U., Bishop,K. D. & James, T. L. (1993). Extracting accurate dis-tances and bounds from 2D NOE exchangeable pro-ton peaks. J. Am. Chem. Soc. 115, 1590±1591.

Liu, Y., Zhao, D., Altman, R. & Jardetzky, O. (1992). Asystematic comparison of three structure determi-nation methods from NMR data: dependence uponquality and quantity of data. J. Biomol. NMR, 2,373±388.

London, R. E. (1990). A theoretical evaluation of the sig-ni®cance of scalar relaxation in coupled spin-1

2 sys-tems in macromolecules. J. Magn. Reson. 86, 410±415.

Ludvigsen, S. & Poulsen, F. (1992). Positive f-angles inproteins by NMR spectroscopy. J. Biomol. NMR, 2,227±233.

Ludvigsen, S., Anderson, K. V. & Poulsen, F. M. (1991).Accurate measurement of coupling constants fromtwo-dimensional nuclear magnetic resonance spec-tra of proteins and determination of f-angles. J. Mol.Biol. 217, 731±736.

MacKerell, A. D. Jr, Nilsson, L., Rigler, R. & Saenger, W.(1988). Molecular dynamics simulations of ribonu-clease T1: analysis of the effect of solvent on thestructure, ¯uctuations and active site of the freeenzyme. Biochemistry, 27, 4547±4556.

MacKerell, A. D. Jr, Nilsson, L., Rigler, R., Heinemann,U. & Saenger, W. (1989). Molecular dynamics simu-lations of ribonuclease T1: comparison of the freeenzyme and 20GMP-enzyme complex. Proteins:Struct. Funct. Genet, 6, 20±31.

MacKerell, A. D., Jr, Rigler, R., Hahn, U. & Saenger, W.(1991). Thermodynamic analysis of the equilibriumassociation and dissociation of 20GMP and 30GMPwith ribonuclease T1 at pH 5.3. Biochim. Biophys.Acta, 1073, 357±365.

Malin, R., Zielenkiewicz, P. & Saenger, W. (1991). Struc-turally conserved water molecules in ribonucleaseT1. J. Biol. Chem. 266, 4848±4852.

Marion, D. & Bax, A. (1988). Baseline distortion in real-Fourier-transform NMR spectra. J. Magn. Reson. 79,352±356.

Marion, D. & WuÈ thrich, K. (1983). Application of phasesensitive two-dimensional correlated spectroscopy(COSY) for measurement of 1H-1H-spin, spin-coup-ling constants in proteins. Biochem. Biophys. Res.Commun. 113, 967±974.

Marion, D., Ikura, M., Tschudin, R. & Bax, A. (1989).Rapid recording of 2D NMR spectra without phasecycling. Application to the study of hydrogenexchange in proteins. J. Magn. Reson. 85, 399.

Martinez-Oyanedel, J., Choe, H.-W., Heinemann, U. &Saenger, W. (1991). Ribonuclease T1 with free recog-nition and catalytic site: crystal structure analysis at1.5 AÊ resolution. J. Mol. Biol. 222, 335±352.

Menke, G. (1984). NMR-spektroskopische Untersuchun-gen zur Struktur der Ribonuclease T1. PhD thesis,University of Frankfurt, Germany.


JMB MS 2812 [3/2/97]

Nanzer, A. P., van Gunsteren, W. F. & Torda, A. E.(1995). Parametrization of time-averaged distancerestraints in MD simulations. J. Biomol. NMR, 6,313±320.

Nishikawa, S., Morioka, H., Kim, H. J., Fuchimura, K.,Tanaka, T., Uesugi, S., Hakoshima, T., Tomita, K.,Ohtsuka, E. & Ikehara, M. (1987). Two histidineresidues are essential for ribonuclease T1 activity asis the case for ribonuclease A. Biochemistry, 26,8620±8624.

Norwood, T. J. (1993). Measurement of the scalar coup-ling and transverse relaxation time of doublets.J. Magn. Reson. ser. A, 101, 109±112.

Norwood, T. J. & Jones, K. (1993). Relaxation effects andthe measurement of scalr coupling constants usingthe E.COSY experiment. J. Magn. Reson. ser. A. 104,106±110.

Olejniczak, E. T., Dobson, C. M., Karplus, M. & Levy,R. M. (1984). Motional averaging of proton nuclearOverhauser effects in proteins. Predictions from amolecular dynamics simulation of lysozyme. J. Am.Chem. Soc. 106, 1923±1930.

Otting, G., Widmer, H., Wagner, G. & WuÈ thrich, K.(1986). Origin of t1 and t2 ridges in 2D NMR spectraand procedures for suppression. J. Magn. Reson. 66,187±193.

Pace, C. N., Heinemann, U., Hahn, U. & Saenger, W.(1991). Ribonuclease T1: structure, function andstability. Angew. Chem. 103, 351±468.

Pearlman, D. A. (1994a). How well do time-averaged J-coupling restraints work?. J. Biomol. NMR, 4, 279±299.

Pearlman, D. A. (1994b). How is an NMR structure bestde®ned? An analysis of molecular dynamics dis-tance-based approaches. J. Biomol. NMR, 4, 1±16.

Pearlman, D. A. & Kollman, P. A. (1991). Are time-aver-aged restraints necessary for nuclear magnetic res-onance re®nement?. J. Mol. Biol. 220, 457±479.

Pfeiffer, S., Engelke, J. & RuÈ terjans, H. (1996). Complete1H, 15N and 13C resonance assignment of ribonu-clease T1: secondary structure and backbonedynamics as derived from chemical shifts. Quart.Magn. Res. Biol. Med. 3, 69±87.

Pople, J. A., McIver, J. W. & Ostlund, N. S. (1968). Selfconsistent pertubation theory. II. Nuclear spin coup-ling constants. J. Chem. Phys. 49, 2965±2970.

Porter, R., O'Connor, M. & Whelan, J. (1983). Editors ofMobility and Function in Porteins and Nucleic AcidsCiba Foundation Symposium 93, Pitman, London.

Post, C. B., Meadows, R. P. & Gorenstein, D. G. (1990).On the evaluation of interproton distances of three-dimensional structure determination by NMR usinga relaxation matrix analysis. J. Am. Chem. Soc. 112,6796±6803.

Press, W. H., Flannery, B. P., Teukolsky, S. A. &Vetterling, W. T. (1989). Numerical Recipes, Cam-bridge University Press, Cambridge.

Quaas, R., McKeown, Y., Stanssens, P., Frank, R.,BoÈckler, H. & Hahn, U. (1988). Expression of thechemically synthesized gene for ribonuclease T1 inEscherichia coli using a secretion cloning vector. Eur.J. Biochem. 173, 617±622.

Richardson, J. S. (1981). The anatomy and taxonomy ofproteins structure. Advan Protein Chem. 34, 167±339.

RuÈ terjans, H. & Pongs, O. (1971). On the mechanism ofaction of ribonuclease T1. nuclear magnetic reson-ance study of the active site. Eur. J. Biochem. 18,313±318.

RuÈ terjans, H., Witzel, H. & Pongs, O. (1969). Protonmagnetic resonance studies on ribonuclease T1. Bio-chem. Biophys. Res. Commun. 37, 247±253.

Saenger, W. (1991). Structure and catalytic function ofnucleases. Curr. Opin. Struct. Biol. 1, 130±138.

Scheek, R. M., Torda, R. E., Kemmink, J. & vanGunsteren, W. F. (1991). In Computational Aspects ofthe Study of Biological Macromolecules by NMR (Hoch,J. C., Poulsen, J. M. & Red®eld, C., eds), PlenumPress, New York.

Schmidt, J. M. (1990). Ermittlung der TertiaÈ rstruktur vonProteinen mit 2D-NMR-Verfahren und Molekular-Dynamik-Rechnungen. PhD thesis, University ofFrankfurt, Germany.

Schmidt, J. M., ThuÈ ring, H., Werner, A., RuÈ terjans, H.,Quaas, R. & Hahn, U. (1991). Two-dimensional 1H,15H-NMR investigation of uniformely 15N-labelledribonuclease T1. Eur. J. Biochem. 197, 643±653.

Schmitz, U., Kumar, A. & James, T. L. (1992). Dynamicinterpretation of NMR data: Molecular dynamicswith weighted time-averaged restraints and ensem-ble R-factor. J. Am. Chem. Soc. 114, 10654±10656.

Schwalbe, H., Rexroth, A., Eggenberger, U., Geppert,T. & Griesinger, C. (1993). Measurement of C0, Cg

coupling constants in 13C labeled proteins: a newmethod for the assignment of g-methyl groups invaline residues. J. Am. Chem. Soc. 115, 7878±7879.

Steyaert, J. & Wyns, L. (1993). Functional interactionamong the His40, Glu58 and His92 catalysts of ribo-nuclease T1 as studied by double and triplemutants. J. Mol. Biol. 229, 770±781.

Steyaert, J., Hallenga, K., Wyns, L. & Stanssens, P.(1990). Histidine-40 of ribonuclease T1 acts as basecatalyst when the true catalytic base, glutamaticacid-58, is replaced by alanine. Biochemistry, 29,9064±9072.

Stonehouse, J., Shaw, G. L., Keeler, J. & Laue, E. D.(1994). Minimizing sensitivity losses in gradientselected 1H-15NHSQC spectra of proteins. J. Magn.Reson. Series A, 107, 178±184.

Thomas, P. D., Basus, V. & James, T. L. (1991). Proteinsolution structure determination using distancesfrom two dimensional nuclear Overhauser effect ex-periments: effect of approximations on the accuracyof derived structures. Proc. Natl Acad. Sci., USA, 88,1237±1241.

Torda, A. E., Scheek, R. M. & van Gunsteren, W. F.(1989). Time-dependent distance restraints in mol-ecular dynamics simulations. Chem. Phys. Letters,157, 289±294.

Torda, A. E., Scheek, R. M. & van Gunsteren, W. F.(1990). Time-averaged nuclear Overhauser effectdistance restraints applied to tendamistat. J. Mol.Biol. 214, 223±235.

Torda, A. E., Brunne, R. M., Huber, T., Kessler, H. &van Gunsteren, W. F. (1993). Structure re®nementusing time-averaged J-coupling restraints. J. Biomol.NMR, 3, 55±66.

van Gunsteren, W. F. & Berendsen, H. J. C. (1987). Gro-ningen Molecular Simulation (GROMOS) Library Man-ual, Biomos, Groningen.

Vuister, G. W. & Bax, A. (1992). Resolution enhance-ment and spectral editing of uniformely 13Cenriched proteins by homonuclear broadband 13Cdecoupling. J. Magn. Reson. 98, 428±435.

Wang, A. C. & Bax, A. (1995). Reparametrisation of theKarplus relation for the 3J(Ha-N) and 3J(HN-C0) inpeptides from uniformly 13C/15N enriched humanubiquitin. J. Am. Chem. Soc. 117, 1810±1813.


JMB MS 2812 [3/2/97]

Wider, G., Neri, D., Otting, G. & WuÈ thrich, K. (1989). Aheteronuclear three-dimensional NMR experimentfor measurement of small heteronuclear couplingconstants in biological macromolecules. J. Magn.Reson. 85, 426±431.

Widmer, H., Widmer, A. & Braun, W. (1993). Extensivedistance geometry calculations with different NOEcalibrations: New criteria for structure selectionapplied to sandostatin and BPTI. J. Biomol. NMR, 3,307±324.

Williamson, M. P., Havel, T. F. & WuÈ thrich, K. (1985).Solution structure of proteinase inhibitor IIa frombull seminal plasma by 1H HMR and distancegeometry. J. Mol. Biol. 182, 295±315.

Withka, J. M., Srinivasan, J. & Bolton, P. H. (1992). Pro-blems with, and alternatives to, the NMR R factor.J. Magn. Reson. 98, 611±617.

WuÈ thrich, K. (1986) . In NMR of Proteins and NucleicAcids, pp. 122±125, John Wiley & Sons, New York.

Zhao, D. & Jardetzky, O. (1994). An assessment of theprecision and accuracy of protein structures deter-mined by NMR. J. Mol. Biol. 239, 601±607.

Zegers, I., Haikal, A. F., Palmer, R. & Wyns, L. (1994).Crystal structure of RNase T1 with 30-guanylic acidand guanosine. J. Biol. Chem. 269, 127±133.

Zuiderweg, E. R. P., Mcintosh, L. P., Dahlquist, F. W. &Fesik, S. W. (1990). Three-dimensional carbon-13-resolved proton NOE spectroscopy of uniformlycarbon-13-labeled proteins for the NMR assignmentand structure determination of larger molecules.J. Magn. Reson. 86, 210.

Edited by P. E. Wright

(Received 13 August 1996; received in revised form24 October 1996; accepted 29 October 1996)

Supplementary material, comprising one Table, isavailable from JMB Online.


JMB MS 2812 [3/2/97]

Limits of NMR Structure Determination using …brd/Teaching/Bio/asmb/current/...Limits of NMR Structure Determination using Variable Target Function Calculations: Ribonuclease T1,a

Documents