00:00:00 Where a factor of a hundred more computer time power would be very useful in

00:00:05 Letting us really understand on a nanosecond timescale how protein active sites relax on response to different ligands

00:00:13 So there may be other you know algorithmic improvements, which helped us along

00:00:17 But I think this is an area where a couple or more orders of magnitude would be very useful. Thank you Peter

00:00:23 We now have Herschel on line 5. Herschel?

00:00:27 Hello Peter, this is Herschel Weintraub at Merrill Dow in Cincinnati

00:00:31 There's been some controversy that variations in

00:00:35 Windowing and other parameters will allow prediction of any either sign or value at the Delta Delta G

00:00:41 If this is the case that would limit the predictive value of the method although it would still allow

00:00:47 Retrospective analysis is this a result of the inability to calculate a small enough window or a long enough?

00:00:55 Simulation or is it something inherent in a instability of the technique?

00:01:00 That's a very good question Herschel nice to hear from you

00:01:03 the I think that with suitably chosen potential functions one can get very

00:01:09 Reliable numbers as I try to illustrate particularly for polar molecules reproducible numbers

00:01:15 and

00:01:16 And for ionic systems also with with the cutoff correction one gets very reproducible results if one can run the simulation

00:01:24 For let's say a hundred picoseconds

00:01:27 however for systems where there are large van der Waals changes

00:01:32 Then you may have to run into the nanoseconds for instance for

00:01:37 Mutating methane to nothing in water to get your answers to converge to tens of kilocalories

00:01:42 So a lot of your your things that you've heard I agree with some of it

00:01:47 But not with others in other words if you're happy with one kilocalorie

00:01:50 I think the current methodologies on solvation can do this for many molecules

00:01:55 All right, I believe you also have a question for Peter

00:01:58 Well, it's it's related to the whole problem of how much time does a computation really take?

00:02:06 We've talked about the computer time, but really how much time does it take?

00:02:11 How much time do you does a graduate student or somebody have to devote to a calculation?

00:02:16 To you know produce a reasonable answer that they're

00:02:21 That they're willing to publish I guess right that's a good question

00:02:24 And let me it gives me the opportunity to bring it to relate some of what my remarks to what Bill Jorgensen?

00:02:30 alluded to with base pairing in water study that was done by Liam Dang and in which we

00:02:36 calculated the absolute free energy of association of

00:02:40 Base pairs either stacked or hydrogen bonded in water an AT base pair

00:02:45 so we did the complementary simulation to what Bill had done in chloroform and

00:02:49 Confirmed what he thought would happen in that

00:02:53 It the stack form was more stable the hydrogen bonded one is virtually not now that simulation took of the order of

00:03:02 You know took hundreds of hours of Cray time because one had to set sample so much carefully

00:03:08 Carefully the association, and we're not sure we even in that case we sampled enough possible confirmations of base stacking

00:03:16 Although we know it's more stable because that kind of error would only increase the stability of the stacked form

00:03:23 But that calculation took I would say you know six months of sort of setup time thinking

00:03:30 Calculation on Liam Dang's part, and he is a very experienced modeler

00:03:34 So many of these calculations take a long time and careful thought I mean most of them do

00:03:39 Okay

00:03:40 We have a caller on line six Jan or Ben. Would you like to state your question, please hello?

00:03:46 I would like to know how you decide your starting confirmations

00:03:51 While you make the complex with the enzyme?

00:03:55 I mean to say how do you orient your inhibitors with the binding site and

00:04:02 How do you select the starting confirmation?

00:04:06 That's a very good question

00:04:08 The what we have done so far the two examples. I showed you on thermo lysin and

00:04:14 Ribonucleic nuclease t1 we had the structure of an inhibitor in each case

00:04:19 In the case of ribonuclease we didn't have the structure of the AMP model

00:04:24 And we created that by mutating the GMP to the AMP so that's a problem that

00:04:30 Combined use of what Jeff Blaney has done coming up with if one doesn't have the structure of a complex that is using distance geometry

00:04:37 And perhaps the second technique he he mentioned that was so successful on

00:04:44 methotrexate

00:04:45 DHFR interactions using that as a starting guess and then doing the mutation with with a few

00:04:51 Representative confirmations that come out of that guess if one doesn't have the structure of the crystal structure of a complex

00:04:57 So some of those distance geometry methods might be useful if one doesn't have a starting geometry for the protein inhibitor complex

00:05:05 Peter we have penny on line six penny your question

00:05:11 Can you hear me?

00:05:13 Line nine are you there?

00:05:15 short time span

00:05:17 Individual motions is there any effort to use such methods to look at large-scale?

00:05:24 Cooperative motions of larger pieces of a molecule

00:05:30 Yes, there are such efforts the time the limitation as I

00:05:35 Alluded to in my talk is that in molecular dynamics one can take only very the standard molecular dynamics one can take only very small

00:05:43 time steps

00:05:45 femtosecond and thus one is limited to

00:05:48 More short timescales just by computer limitations. There are attempts to develop new algorithms for instance by

00:05:56 Tamar schlick of New York University has another algorithm for doing molecular dynamics that might allow 10 to the minus 13th

00:06:03 second time steps, but these are areas of constant

00:06:08 Development attempting to make them the dynamics algorithm more efficient and allowing us to simulate larger and larger changes

00:06:16 Bonnie on line eight your question

00:06:19 Yes, I'm Bonnie Gun from Oxford, Mississippi

00:06:22 I'm interested you mentioned the fact that the free energy is greater in the salvation

00:06:28 So then the hydrogen bonds your stacking energy is being greater is the base stacking energy for the free for the new

00:06:36 They repeat this is the base stacking free energy and nucleic acids greater than that

00:06:41 Which are present in hydrogen bonds in solution if one applies these principles to polymeric DNA

00:06:50 That's a good question I think in polymeric DNA again coming back to what Bill Jorgensen said

00:06:56 We've got these ionic

00:06:57 Salvation problems and we have all the complexity of the other bases stacking on top of the the one base

00:07:03 And I think we haven't sorted that out yet

00:07:05 I think that's a an issue that's going to be sorted out in more complex DNA the role of hydrogen bonding and base stacking

00:07:12 but our preliminary analysis based on molecular mechanics and experimental results by people like Doug Turner of

00:07:18 Rochester University

00:07:20 Suggests that both play a critical role in DNA and RNA stability

00:07:26 Gentlemen, we need to break this up and we'll come back to it after we hear from Dr. David Case. Peter

00:07:31 We'll see you one more time for our panel discussion. Thanks very much

00:07:35 Dr. Case is a colleague of yours Art

00:07:38 Yes, Whitney. Dave has been a member in the Department of Molecular Biology at Scripps Clinic since 1986

00:07:44 He's been applying molecular modeling techniques to answer a number of biological questions

00:07:48 Including simulating the dynamics of oxygen escaping from myoglobin and hemoglobin

00:07:53 Today Dave is going to talk about the work. He's been doing tying molecular dynamics to experimental data

00:07:59 So here is Dr. Case with determining three-dimensional solution structures of proteins from NMR data

00:08:09 Thank you Whitney

00:08:11 As Art indicated one of the most promising areas of application of molecular modeling

00:08:16 Involves a close combination of theory and experiment in which the observed data serves as constraints in a molecular dynamics simulation

00:08:23 This powerful combination allows the modeler to make a fairly wide survey of conformational space

00:08:29 Looking for conformations that satisfy experimental constraints and at the same time have low computed energies

00:08:35 Two prominent sources of experimental data are x-ray crystallography and NMR spectroscopy

00:08:41 Essentially the same methods can be used in both fields and I will be talking about the ways in which

00:08:46 Dynamical simulations can be used in conjunction with NMR to determine the three-dimensional solution structures of proteins

00:08:54 The basic outline of the procedure is shown in the first slide

00:08:58 One begins with sequence specific assignments that is

00:09:02 Identifications of peaks in the NMR spectrum with particular protons in the sample

00:09:07 The ways in which these assignments are made are fascinating in their own right, but are outside the scope of my talk today

00:09:14 As I will illustrate briefly

00:09:16 Cross peaks in these spectra enable one to assign distance constraints that the structure must satisfy

00:09:22 Since cross peaks will appear only when two protons are within about five angstroms of each other

00:09:27 the theoretical procedure involves the use of distance geometry with which Jeff Blaney has already talked about and

00:09:34 Restrained molecular dynamics simulations the final structures can approach the quality of crystallographic structures

00:09:41 I'll be using as an example the zinc finger structure recently solved in a collaboration

00:09:46 Between Peter Wright's NMR lab at Scripps Clinic and my group

00:09:50 Zinc fingers are a recently discovered class of DNA binding domains

00:09:54 Each domain is a short stretch of polypeptide that has four side chains bound to zinc

00:10:00 The regulatory protein may contain many such domains

00:10:03 Shown in here is the sequence of the peptide that we studied

00:10:07 Which is a member of the class of zinc fingers that bind the metal through two cysteine and two histidine side chains

00:10:14 Our peptide is 25 amino acids in length and the circled residues are conserved or nearly conserved in all members of this class

00:10:23 Although we studied a synthetic peptide with this sequence domains like this appear repeatedly in real regulatory proteins

00:10:31 This slide shows a portion of the nuclear overhauser effect spectrum for the zinc finger peptide

00:10:37 The amide amide region is shown here and later on. We'll see the amide in this slide

00:10:42 We see the amide alpha proton cross peaks peaks along the diagonal of these two-dimensional spectra correspond to unique

00:10:49 identified protons in the spectrum

00:10:51 The off-diagonal peaks several of which are labeled here

00:10:55 Indicate pairs of protons that must be close in space since magnetization has been transferred between them during the course of the experiment

00:11:03 About a hundred and fifty such cross peaks have been measured for the zinc finger

00:11:07 For a larger protein say of a hundred amino acids it can be possible to identify more than a thousand such peaks a

00:11:15 Summary of the short-range peaks is shown here

00:11:18 The boxes across the top show sequential cross peaks that is cross peaks between the amide or C

00:11:25 Alpha-proton of residue I and the amide proton of residue I plus one

00:11:30 Peaks like this are diagnostic of the local secondary structure

00:11:34 Typically alpha amide cross peaks are seen on extended structures like sheets and amide amide peaks occur in helices

00:11:42 further evidence for helix in the region of residues 13 to 22 is the existence of I to I plus 2 I

00:11:49 To I plus 3 and I to I plus 4 cross peaks which are shown as bars below the boxes

00:11:55 In many instances one can simply read off helix and sheet structure from plots like this without doing any calculations at all

00:12:03 The distribution of longer range cross peaks is shown here

00:12:08 These are the peaks that eventually determine how the protein secondary structure elements interact with each other to give the tertiary structure of the protein

00:12:17 In a peptide as small as the zinc finger only a dozen or so such peaks are seen

00:12:22 But we shall see that these are sufficient to determine the overall structure and again for larger proteins

00:12:28 The number of observed long-range peaks is of course much larger

00:12:32 The experimental input then consists of a number of distance constraints to divide from the nuclear overhauser effect data

00:12:39 In addition as Jeff talked about we know a lot of information about distances just from knowing the chemical structure of the protein

00:12:47 Turning distances into three-dimensional structures is generally accomplished through distance geometry calculations

00:12:53 The basic mathematical ideas are reviewed in the next slide

00:12:57 The metric matrix G is defined to give the dot product between the coordinate vectors of two atoms

00:13:03 Of course, we don't know the coordinates if we don't know the structure

00:13:07 But from the law of cosines the dot products can be computed from the distances. I

00:13:12 Don't show it on this slide, but it is also the case that distances from the center of gravity position

00:13:17 Oh can also be determined just from inner atomic distances

00:13:22 The net result is that the metric matrix G can be determined if you know all of the inner atomic distances

00:13:29 Since only some of these are known one has to make a guess for the remainder

00:13:33 Repeatedly running through the algorithm with different guesses then generates a variety of structures

00:13:40 Once the metric matrix is available three-dimensional coordinates can be obtained by the method outlined in the next slide

00:13:46 This slide shows a theorem which states that the metric matrix must be of rank 3 in order to identify with a real object

00:13:54 that means that the metric matrix G must have just three nonzero eigenvalues if

00:14:01 Lambdas are the eigenvalues and W is the eigenvector then the relation boxed at the bottom of this slide

00:14:07 Allows one to construct the coordinates X from these

00:14:11 If the metric matrix G actually has more than three nonzero

00:14:15 Eigenvalues one can simply ignore all but the first three and still use this formula

00:14:21 This procedure amounts to projecting the object defined by the distances on to three-dimensional space

00:14:28 The structures directly resulting from distance geometry are generally quite poor by NMR standards

00:14:34 Partially because not all of the distances are known and partially because of the loss of information

00:14:39 Entailed in throwing away all but the first three eigenvalues

00:14:44 We typically use restrained molecular dynamics to improve or refine these structures

00:14:49 in this procedure the energy function used is a combination of a molecular mechanics force field and

00:14:55 Penalty functions that are defined to become increasingly positive as the NMR constraints are violated

00:15:02 One could just perform an energy minimization in this combined potential

00:15:06 but energy minimization tends to get trapped very easily in local minima a

00:15:10 More robust optimization procedure uses molecular dynamics to anneal the system as illustrated in the next slide

00:15:18 The heavy line in this slide shows the system temperature as a function of simulation time for the protocol that we most often use

00:15:26 In a vacuum calculation there is no problem with heating the system to 1200 degrees Kelvin

00:15:31 Especially since the presence of many NMR constraints forces the system to stay near its experimental geometry

00:15:37 what the high temperature equilibration does is allow the system to explore many regions of conformational space as

00:15:44 The temperature is slowly reduced back to zero most of the molecule most often will fall into a relatively good local minimum

00:15:52 Although this is certainly not a global minimization algorithm

00:15:55 We have found that the more slowly you can afford to cool the system the better on the average the results will be

00:16:01 Although one quickly hits a law of diminishing returns where further simulations don't lead to much improvement

00:16:08 How good are the structures that one obtains by this method?

00:16:12 This is a difficult question to answer since the true solution structures are not known

00:16:17 One approach is to look for convergence of structures for different choices of inter atomic distances

00:16:22 If randomized starting points all lead to the same final structure one might have some

00:16:27 Confidence that these are being driven by experimental data. That is that the resulting structures are quote real unquote a

00:16:35 Second approach to judging quality is to compare the results with x-ray crystallographic structures where these are available

00:16:41 I'll give an example of each

00:16:44 The first example is of the zinc finger

00:16:46 I mentioned earlier where the NMR spectra were taken by Min Lee and Peter Wright at Scripps Clinic and the computer analysis was done by

00:16:53 Dr. Soman, Gary Gifford and myself

00:16:56 The overall structure of the zinc finger is shown here

00:16:59 there is an extended portion and a turn at the end terminal end at the top of the video and

00:17:04 A fairly long helix toward the C terminal end at the bottom

00:17:08 the zinc atom is bound to two cysteines that are on opposite sides of a reverse turn and

00:17:13 To two histidines that are a part of the long helix

00:17:16 you first saw a ribbon representation of the polypeptide backbone backbone and now a space-filling view of the entire molecule as

00:17:24 An aside this was this video was recorded in real time on a Sun workstation with a graphics accelerator board

00:17:30 And gives you a feeling of the interactive rendering power of relatively inexpensive equipment

00:17:36 The next slide shows a superposition of 37 structures for the zinc finger peptide

00:17:41 The backbone is shown in orange along with the yellow cysteine and blue histidine ligands to the zinc

00:17:47 It is clear that the backbone and ligation geometries are quite well determined

00:17:51 Some statistics about this structure are shown in this figure

00:17:55 Which shows the standard deviations in the backbone first to the angle Phi and in the next slide of the angle Psi

00:18:02 In the turn region from residues four to eight and at the C terminus there is considerable uncertainty in the backbone structure

00:18:08 But the backbone confirmation in the helical region from reach from residues 13 to 22 is very well described

00:18:15 In the next slide I add side chains to the picture of the molecule

00:18:19 You can see that there is much more uncertainty about the positions of side chains than there was about the polypeptide backbone

00:18:26 Still some interesting conclusions can be drawn

00:18:29 Possibly such charge side chains are shown in blue

00:18:31 And it is clear in spite of the spread in the structures that one side of the long helix along the right

00:18:37 has an array of positively charged side chains that could be well situated to interact with a groove of DNA an

00:18:44 Artist rendering of what might be happening is shown in the next slide

00:18:47 Where here you are looking down the long helix of the peptide as it is positioned in the major groove of DNA I?

00:18:54 Should emphasize that we have no clear experimental evidence for this mode of binding although further experiments are in progress

00:19:01 It is also worth remembering that the actual regulatory proteins had many such domains

00:19:05 And that is likely the binding to DNA involves cooperative interactions among these domains

00:19:12 As a second example, I'd like to talk briefly about plastocyanin, which is an electron transport protein active in photosystem one

00:19:21 Here the NMR spectra were provided by John Moore and Peter Wright in my department and the computational work was done by Gary Gifford and myself

00:19:29 This protein has 99 amino acids and is more of a real protein than is the zinc finger

00:19:34 And we can compare our solution structures with crystallographic results from a slightly different plastocyanin

00:19:40 The NMR structure uses the protein from French bean leaves whereas the x-ray structure is of the poplar leaf protein

00:19:47 But the two are expected to be very close

00:19:50 The second video clip shows the NMR structure in blue and the x-ray structure in yellow

00:19:56 in addition to the backbone tube you will see a copper atom and it's for ligands at the top of the molecule and

00:20:03 For phenylalanine side chains that pack closely to form a hydrophobic core in the center of the protein

00:20:09 In this case we have much more extensive data than is currently available for the zinc finger

00:20:14 And we are able to pinpoint the locations of the side chains with greater confidence

00:20:18 I think you will agree that there was a remarkable agreement between the solution and crystal structures as you saw

00:20:24 That last animation was not made in real time, but consisted of 720 frames that differed by a rotation about the vertical axis

00:20:32 It took us about 12 hours to render the images and about a day to record them one by one on the videotape

00:20:40 One important difference between the data set for plastocyanin and that for the zinc finger is our knowledge in plastocyanin of

00:20:47 Stereospecific assignments for many of the methylene beta protons and for the methyl groups in valine and leucine

00:20:53 That is in favorable circumstances the two pro chiral protons on the c-beta side chain can be individually assigned

00:21:01 This allows one to generate much more specific and tighter experimental constraints and the next slide

00:21:07 I show a portion of plastocyanin, which was calculated from structures before such stereospecific assignments were available

00:21:14 That data set is about the same quality as I talked earlier with for the zinc finger

00:21:20 As you can see most of the side chains are poorly described with large uncertainties in their positions

00:21:25 and this again is roughly comparable to the results we found for the zinc finger and

00:21:29 Arises largely from a lack of cross peaks that could help restrict the values of the side chain dihedral angles

00:21:35 the next figure shows the same structures in red that you saw in a previous slide and

00:21:41 Superimposed on top of them structures determined with additional constraints the spread in position has now been substantially reduced

00:21:48 furthermore the new positions closely resemble those seen in the x-ray structure, which are shown in light blue a

00:21:54 Similar figure is shown next for the four central phenylalanine residues

00:21:59 Again, the red structures were determined with an early data set and the yellow with more complete information

00:22:05 The newer structures are better defined and in general closer to the x-ray configurations

00:22:10 This is especially noticeable for phenylalanine 82 at the center left

00:22:18 Aside from just applying this general procedure to new proteins

00:22:21 There are a number of improvements in the refinement procedure that appear to be feasible

00:22:26 My group is looking closely at quantitation of the cross peak intensities

00:22:30 Rather than simply saying that the existence of a cross peak implies a short distance, which is really only a qualitative statement

00:22:37 We are back calculating the nuclear overhauser spectrum from our structures and refining the structures into this data

00:22:44 The basic problem mathematical problem is outlined in this slide

00:22:49 The magnetization transfer process in the experiment involves not just two spins at a time

00:22:55 But actually all of the spins in the protein which interact according to a set of linear kinetic equations the block equations

00:23:02 Which are shown at the top of the slide?

00:23:04 The rate matrix elements are depend upon the inverse sixth power of the distance and also upon the spectral density function J

00:23:12 Which in simple cases has the Lorentzian form shown at the bottom of the slide?

00:23:17 Here tau is the rotational correlation time for a rotational diffusion of the entire protein in

00:23:23 Many cases it appears that this simple model for motion is adequate to understand the motional contribution to the nuclear overhauser spectrum

00:23:31 by solving the differential equation shown at the top of this slide the

00:23:35 Intensity of each peak can be calculated and compared with experiment in a process that is completely analogous to that used in crystallography

00:23:43 where computed and experimental structure factors are also compared as

00:23:47 in crystallography initial structures can be updated to improve agreement with the observed spectrum and

00:23:53 Overall figures of merit such as our factors can be determined to quantify the agreement

00:23:58 This is a very new area, but one of them that offers exciting possibilities for getting additional accurate structural information

00:24:05 from NMR data

00:24:20 While we're waiting for your calls to come through I believe art has a question nice graphics

00:24:28 Actually I wanted to pursue that the last point that you made about actually doing refinement on a figure of merit

00:24:35 analogous to crystallography

00:24:37 It sounds like a great idea especially being a lapse crystallographer to have a number to associate with

00:24:43 The quality of a structure, but where do you think the bottlenecks are in and actually implementing that approach?

00:24:49 Well, that's a very computationally intensive approach right now, and that's one of the bottlenecks

00:24:54 But I think we can actually do it on on proteins of the size of system

00:24:57 We can afford to do NMR on right now

00:24:59 We're seeing

00:25:01 Crystallographic our factors are things that are calculated in the same way as crystallographic our factors in the range of 10 to 20 percent for

00:25:07 Many of our structures. We don't know if that 10 to 20 percent number. It means the same thing as a crystallographic our factor

00:25:14 The big bottleneck for NMR is really that we're still limited to proteins of less than about 150 amino acids

00:25:22 So we can't begin to approach the size of problem yet that

00:25:26 Crystallography can can handle all right we have bill on line 5 bill your question

00:25:32 Yes, this is a question for anyone on the panel

00:25:36 I have a

00:25:38 Group of University of Tennessee pharmacy students with me taking a course in drug design

00:25:44 Could you provide them with any examples of drugs?

00:25:49 In clinical trials or beyond that have been discovered through molecular modeling

00:25:56 Well, I guess I'm the only person up here with it with a microphone on right now

00:26:00 And I don't know the answer to that

00:26:03 I maybe we'll ask Jeff that question during the panel discussion, and we'll save that one

00:26:06 I think that that's a good question to

00:26:09 Get everybody's opinion on

00:26:13 Do you have another question?

00:26:15 well actually I was interested in the general problem of

00:26:20 Getting ensembles of structures out

00:26:22 I mean we're used to looking at a crystal structure and thinking of it as the structure and now

00:26:28 More and more people are beginning to realize that that these this might not be the structure

00:26:33 But part of an ensemble of structures, especially

00:26:36 the solution

00:26:37 structures

00:26:38 How how do you think we're going to deal with?

00:26:42 An explosion of information in one way or another how are we going to represent that when we distribute that kind of data?

00:26:47 Well distributing it. I think just means we have to distribute more

00:26:51 examples from the family

00:26:53 I'm not sure that's any worse of a problem in solution than it is in crystals because most protein crystals

00:26:58 have a lot of water in them and actually all the evidence we have indicates that the

00:27:02 Uncertainty or the spread of the family and crystals is about the same as it is in solution at room temperature

00:27:08 crystallographers are facing up to this problem finally as

00:27:11 NMR spectroscopists are in the sense that that one has to refine a number of structures into the data and

00:27:18 Eventually in places like the protein databank one has to report those to the to the community and then we're still developing those methods

00:27:24 But but I think they're coming all right you have calls waiting the first one is ping on line six your question ping

00:27:31 Yes

00:27:33 This is Ping Yang call from Pima more in Terra Haute, Indiana. I have a question

00:27:40 When you have fast exchange in my proton in your

00:27:45 system

00:27:46 It's very hard to get the NOE data, and do you have method to?

00:27:53 special especially

00:27:55 deal with that problem, or if you don't then do you miss the

00:28:00 distance measurement on that set of data and

00:28:05 The distance geometric calculation method suffer from the point. That's a good question

00:28:11 Clearly it's easiest to do the experiment in d2o

00:28:14 Because you don't have to get rid of the water peak in the NMR spectrum

00:28:17 And so for the protons as you mentioned that don't exchange rapidly

00:28:21 With solvent one can put make the solvent become d2o and do the experiment

00:28:25 But for proteins one can also do these experiments in water solution where all of the amide peaks are there and although

00:28:32 There are problems with the pre saturation of the water pulse

00:28:35 Spectrometers and pulse sequences keep getting better and better and we could actually get all of the amide amide data

00:28:41 We need an amide alpha data and so on from experiments in water as well as experiments in d2o

00:28:46 So it's not as big a problem as as you might have guessed from your question

00:28:50 The problem is not the water exchange is the a my proton it has the fast exchange

00:28:59 Situation not just with the water if the amide proton itself

00:29:03 You will get the overlapping peak on the broke raw peak amide region. It's not that water exchange

00:29:10 Directly, okay. It may have the fast

00:29:13 exchange on the amide proton itself and then in that case you have the difficulty to locate the

00:29:20 position on

00:29:21 the

00:29:23 NOE and

00:29:25 NH proton, okay. I'm not talking about a pre saturation water signal technique itself

00:29:32 Well, certainly the broad

00:29:34 Certainly the broadness of the peaks is a problem both from exchange effects and once as the protein gets bigger and bigger

00:29:41 rotational diffusion slows down and the peaks become broader for that reason too and and it's a real

00:29:47 Science and I'm not really an NMR spectroscopist, but I'm a computer person myself

00:29:51 But but it's it's a real art and science to be able to resolve and quantitate

00:29:56 NOE information off of amide protons as you say and I don't know completely where the limits of that will be

00:30:04 some

00:30:05 Improvement can be can be obtained for example by random partial due duration of large proteins

00:30:10 Frank line eight go ahead with your call. Oh, Dave. This is Frank Brown. I'm with Glaxo and RTP, North Carolina

00:30:18 My question is about the protocol that you outline at the beginning your talk

00:30:23 In the protocol that you have stated you would collect the data and the distances and go on into your structure generation

00:30:30 But how do you know if you have enough data to give you a decent structure as the protocol you outline?

00:30:37 Without enough data, you'd go and make a bunch of structures which become meaningless

00:30:42 Well, I don't think I think in most cases in proteins

00:30:45 You don't get completely meaningless data

00:30:46 but you can see with the less data you have the the wider the spread and the structures is and

00:30:52 And certainly one has to also worry about errors in the data

00:30:55 If there's not a lot of redundant data because then individual mistakes can make a bigger a bigger importance

00:31:00 There also is an iterative effect of these things in the sense that once you have some preliminary structures

00:31:06 You can also often use those structures to to assign more of the of the cross peaks in your spectrum

00:31:11 And therefore you bootstrap your way into more data

00:31:14 Right now we don't really have any way of saying before you do the calculation whether or not you're going to get a good structure

00:31:21 But but it's generally pretty evident from both the number of structures that converge and the divergence amongst the converge structures

00:31:28 Whether or not you have enough data in the end

00:31:31 The examples I showed tended to be for proteins where we do have a lot of data

00:31:35 Things don't look nearly that good when you have a lot less data as you might expect

00:31:40 Line 5 like she me your question, please

00:31:44 This is Lakshmi Narasimhan from Upjohn Company, Kalamazoo, Michigan

00:31:48 Dr. Case, are there any examples of cases where even tight binding inhibitors

00:31:55 their distances to protein atoms have been

00:32:01 Identified through NMR techniques

00:32:05 Yes, there are there's a whole class of experiment called a transferred NOE experiment in which

00:32:11 Essentially you you do the NMR on the this is the case where you have a ligand that's an exchange

00:32:16 It has to be actually fast exchange with

00:32:18 With bound state in a free state and you actually do the NMR on the free state where the assignments are easy

00:32:24 but because of

00:32:25 That those nosy peaks are actually remembering what where they were in the in the bound state

00:32:31 The people have done the most with this or the NIH group of chlorine grown and born their co-workers

00:32:36 And there are a number of examples of this

00:32:39 It won't work in really tight binding situations because then the off rate for getting the ligand off of the protein is too slow

00:32:46 But in a fairly wide range of intermediate situations the transferred NOE experiment

00:32:51 I think will start to tell us a lot about inhibitor binding

00:32:55 even in cases where maybe the

00:32:57 Thing that is binding to as much too large to do the complete NMR experiment on on the protein itself

00:33:03 David we have a call from you on Richard from Richard on line six

00:33:08 Yes, Dr. Case

00:33:10 Many molecules of biological interest are embedded within a lipid matrix such as a membrane

00:33:17 have

00:33:18 structures of this sort studies on structures of this sort been undertaken and

00:33:23 What approaches might be needed to determine?

00:33:27 a structure of a protein within this environment

00:33:30 Yeah, that's a really very good question and and actually the methods I talked about won't work

00:33:35 At least in their present form in situations like that because the high resolution NMR

00:33:41 Which is required to assign all the resonances to get the distance information that I use is

00:33:46 Generally available only when the molecule is tumbling fairly rapidly in a solvent like water or some other similar NMR like solvent

00:33:53 There has been some recent work not necessarily in membranes

00:33:57 But in lipid vesicles at getting some amount of high-resolution NMR

00:34:01 And actually getting enough tumbling time for for molecules that are embedded in in my cells and kind of artificial

00:34:09 nonpolar environments

00:34:11 but the level of precision and detail that one is seeing in those experiments is really reminiscent of

00:34:17 Water NMR ten years ago where you may get a few distances that actually may be very important

00:34:22 But nothing yet like the ability to determine an entire structure in those kinds of environments

00:34:28 All right, Frank is back again on line eight with a follow-up question for you

00:34:35 About the data beforehand you can actually look at the correlation in the data before you actually undergo

00:34:42 minimization or your dynamics to understand whether or not you have a high enough correlation to give you a

00:34:49 Decent refined structure and that's called the template

00:34:53 analysis

00:34:55 The NMR template analysis and it's an available technique that you can use to identify regions, which will have a stable structure

00:35:04 That's right

00:35:04 Frank is referring to some work that was published this year or last year in the Journal of the American Chemical Society and some other

00:35:10 places on

00:35:11 on methods both to handle kind of determining how good the structures might be and also to handle their the very interesting problem of

00:35:19 Deciding in a peptide or some other small system that might be floppy or a drug system

00:35:24 Where you might not even know in advance whether there was a quote native structure or not

00:35:28 Whether it was worth your while to to pursue that or whether what what the data should be expected to to determine it

00:35:35 I'm glad you you brought that up

00:35:37 when I said that you really had to do the calculation first time is maybe looking at a

00:35:41 Situation where it was clear that one had hundreds to thousands of cross peaks and a native structure of a protein

00:35:47 But the real kind of detailed question about the quality of the structure, I don't know of a good way of deciding

00:35:54 Beforehand for that kind of problem right now

00:35:57 Thank you, David

00:35:58 We're going to take a three-minute stretch break right now so that we can set up the stage for one of our panel

00:36:03 Discussions the phones are open and we encourage you to start placing your calls now one last time the numbers to call are

00:36:10 1-800-942-1515 for California

00:36:15 1-800-972-1515 for the US and 1-619-265-6429

00:36:22 For Canada

00:39:30 We have callers waiting the first one is Nathan on line 5. Nathan your question, please

00:39:36 Hello, my name is Nathan Singh from the University of Kentucky

00:39:39 Could someone on the panel, please elaborate on the benefits and drawbacks of various minimization procedures

00:39:46 Such as steepest descent conjugate gradient and Newton Rolfson procedures

00:39:52 Thank you

00:39:54 Let's start out Peter

00:39:55 yeah

00:39:57 the the techniques steepest descent and

00:40:02 conjugate gradient

00:40:04 Are the simpler methods they require less computer memory

00:40:09 But they converge less well near the minimum

00:40:14 The Newton Rafson techniques have the disadvantage in that they require a storage of an n by n matrix where?

00:40:22 n is the number of

00:40:24 Coordinates 3n in terms of the number of atoms

00:40:27 So one tends to use Newton Rafson if one has systems of up to 400 or 500 atoms Dave case has actually done some

00:40:35 Minimizations on large systems like that, but for much larger proteins or other macromolecules

00:40:42 one needs to use conjugate gradient, which is

00:40:45 quite convergent

00:40:47 but is

00:40:49 Doesn't require them the storage

00:40:51 Steepest descent is just a way to get rid of the first really terrible gradients and doesn't converge that well

00:40:57 So one abandons that quickly and goes to conjugate gradient

00:41:01 Jeff on line 6. Let's hear your question

00:41:05 Jeff Noss calling from the Walter Reed Army Institute of Research in Washington DC

00:41:10 There's a variety of commercial micro graphics and micro modeling packages available on the market

00:41:16 And they use different potential fields charm versus amber amber for example

00:41:21 Is there any studies of differences of these force fields on the results that you can get?

00:41:26 using different classes of model compounds nucleic acids proteins

00:41:31 carbohydrates and lipids

00:41:35 Anyone on the panel

00:41:38 The

00:41:40 The most recent sort of study that I know of is one that took place in

00:41:45 1984 although there are many unpublished papers in the literature or on unpublished unpublished works on

00:41:52 on selected classes of molecules

00:41:55 and I think this is an evolving field where

00:41:59 We don't really know the answer and force fields are continually being refined. So

00:42:04 You just need to look I would advise you just to look at the class of molecules you're interested in and

00:42:11 You know and make an evaluation there mm2 and mm3 is of course

00:42:16 Is the best bet for the most refined molecular mechanics force field for organic molecules?

00:42:22 And it's most likely to be correct for those kinds of molecules

00:42:27 Dr. Coleman, we have Serena on line 8 who also has a question for you

00:42:32 Hello, Dr. Coleman

00:42:33 We're interested in modeling nucleic acid systems that are have platinum

00:42:39 Metal ions interacting with the nucleic acids. I was wondering with the amber force field

00:42:46 What is the best methods of introducing metals into the force field and how does one?

00:42:52 determine force constants and the like

00:42:55 The the best method that

00:42:58 one would use in that case for very covalent interaction is to actually put it in as a covalent bond and the specific example you

00:43:05 Mentioned the platinum nucleic acids has already been studied by a number of groups

00:43:10 Kozelka

00:43:12 Petsko Lippard at MIT have published a number of papers. I think around 1987 on exactly that problem using

00:43:19 Using our molecular mechanics software

00:43:21 I think Kozelka has gone back to Paris and done other things and what they find is an equilibrium between

00:43:27 Non-kinked and kinked DNA in the presence of platinum

00:43:31 All right, Dr. Blaney, Surat is on line 11. Surat your question

00:43:36 My question is directed for Dr. Jeff Blaney

00:43:40 It has two parts

00:43:41 Can the information obtained from 2d NMR be used for docking of this small ligand molecule?

00:43:48 onto a receptor or DNA

00:43:50 If yes, then how do you feed the information of intermolecular context detected by 2d NMR?

00:44:00 All right

00:44:01 The answer is yes, you could use 2d NMR information if you had it for example if you had

00:44:07 Intermolecular NOEs that told you something about distances between specific

00:44:12 Atoms in the ligand and and your protein site you can express those as distance constraints and use them to guide the docking

00:44:20 Other experiments that can give you that beyond the transfer NOE that the Dave mentioned

00:44:25 You might look in papers published by Steve Fessick's lab in which they've done a series of isotope edited experiments

00:44:32 For example with n15 labeled peptides or even n15 labeled protein to get intermolecular NOEs out

00:44:39 All right, dr. Jorgensen you have a question on line 7 from Bonnie you're on Bonnie your question

00:44:49 Are you there Bonnie if so pick up the phone all right perhaps she'll get back to us

00:44:55 But art I believe you had a follow-up comment

00:44:57 All right, actually I wanted to get back to the question that was asked of Dave case during his talk

00:45:02 Which is where are the examples? What are they the examples of?

00:45:06 Say drugs or or other biologically active molecules designed by these methods and I'd like to put it open to the panel

00:45:14 But I'd also like to to open it up to the the callers if any of you out there

00:45:19 Know of and can talk about any of these

00:45:23 success stories call in and tell us

00:45:26 Jeff you're probably the first person to I can mention the examples

00:45:30 I know of two of them are in the references and in my section at the back

00:45:34 The only one I know of that's actually I believe gone as far as clinical trials was the work from welcome on the anti-sickling agents

00:45:40 I don't know how far they've gone with those compounds except. I believe they did make it into the clinic the

00:45:46 Work on the phospholipase a2 inhibitors at DuPont is still going on, but has not made it to clinical trials as yet

00:45:52 There are I believe several examples in the agricultural chemical area and pesticide or herbicide design

00:46:00 And I think if you look for a review article by Toshio Fujita, you may find examples on use of QSAR

00:46:07 That was successfully used to design compounds that I believe may have even gone into commercial production there

00:46:13 There are a variety of other examples Lee Kuypers work at welcome on

00:46:17 using modeling to design better inhibitors

00:46:20 But they have not turned out to be

00:46:23 Compounds useful in the clinic. All right. Dr. Jorgensen. We have Bonnie back on the line on line five

00:46:29 Bonnie

00:46:32 I'm sorry line eight Bonnie

00:46:35 This is Bonnie Gun from Oxford, Mississippi

00:46:38 It has been suggested that the bifurcated hydrogen bond themed in the AT region of a of DNA is

00:46:46 The reason for the stability which has been seen in these melting behaviors

00:46:50 Could these hydrogens bond be in fact strong enough to account for this behavior?

00:47:00 I

00:47:02 Think the

00:47:06 I'm not sure exactly how that relates to some of the results

00:47:09 I I showed and I'm not as familiar with with the I guess the topic as I'd like to be my impression

00:47:17 Is that AT and GC mismatches don't show huge variations in some oligonucleotides?

00:47:25 in melting

00:47:27 Behavior which is surprising in the sense of GC is much has much more

00:47:34 strong hydrogen bonding

00:47:36 on the other hand as a key point of my talk was that you have to keep in mind all of the other effects and

00:47:44 including the solvent effect on these

00:47:47 Interactions, which is very strong and damps out the strong GC hydrogen bonding as compared to

00:47:54 AT in water that we don't see in solvents such as chloroform

00:48:01 All right, we have bill on line one to any of our panelists bill your question

00:48:06 Yes, my question is directed to Peter Coleman

00:48:11 Peter this is Bill Welsh from St. Louis

00:48:14 I enjoyed your program very much

00:48:17 my question relates to

00:48:19 Activity in our group involving metalloproteins and

00:48:23 Could you comment on?

00:48:27 When you're conducting mutations say of one metal ion to another and in particular

00:48:33 Say one metal ion to another one of a different balance state like Fe 2 to Fe 3

00:48:38 What particular problems should one be watchful of in dealing with?

00:48:43 In dealing with

00:48:45 Say perturbation calculations or MD calculations in general regarding such a transformation and number two if I could just interject one

00:48:53 more question

00:48:55 in terms of

00:48:57 parameterizing for metal ions contained in metalloproteins

00:49:02 Could you comment on as you know, there's a lack of of

00:49:06 Good parameters for metal ions in these environments. Could you comment on how one best obtains?

00:49:14 reliable

00:49:15 Force field parameters for metal ions in these situations. Thank you

00:49:20 Nice nice to hear from you Bill

00:49:23 the your question is a very good and difficult one to answer in a general way, but

00:49:30 Kenny Murs, for instance, we worked very hard on zinc parameters

00:49:35 Zinc has a charge of plus two as soon as you get to something that has a charge of more than plus one

00:49:39 It is a great challenge to develop

00:49:43 Molecular mechanical parameters for them and you can use two approaches one is the one I mentioned for the platinum

00:49:49 compounds where you use explicit covalent bonds and don't allow a lot of flexibility and the second is to put a

00:49:56 Formal charge and use a non-bonded interaction to represent the metal which can lead to very large

00:50:02 Electrostatic fields and often give you artifacts. We had to struggle with this a lot in both our thermal isin

00:50:08 Calculations and the ones that Kenny Murs has done on carbonic anhydrase

00:50:12 So I guess I'd have to be fairly wishy-washy here and say you use as much

00:50:16 Experimental data as you can and you test out your model for instance to make sure it's behaving correctly on small model systems

00:50:23 Before you go on to more complicated systems

00:50:26 But this is an example of a kind of thing that Bill Jorgensen

00:50:30 Answered with respect to ionic systems doing the free energy calculations on those is a tremendously

00:50:36 Challenging and difficult thing because there are such large energies and there's so large cutoff corrections. Can I just add something to that Peter?

00:50:43 One one thing to avoid or to recognize is that one transition metal ions, especially are in are in ligation complexes

00:50:51 The

00:50:52 Concentration of charge on the metal ion is not anyplace nearly as large as you might think from the formal oxidation stage

00:50:57 And one generally has to do some quantum mechanical calculation to get a distributed charge model

00:51:02 So that in going from iron 2 to iron 3 one doesn't change just one electron charge on the metal

00:51:08 But the but the charge is almost always distributed over quite a large number of atoms

00:51:12 Jeff it's just one other comment that I had on that

00:51:16 Donuts has published at least two papers that I can think of on a force field. They've developed for doing metalloproteins

00:51:23 This has been published sometime in the last few years

00:51:27 Dick line one go ahead, please

00:51:31 Dick Keys Cal State LA this is a general question to any of the panelists

00:51:38 The question is whether there's enough experience with these systems to apply

00:51:46 expert systems to looking for

00:51:49 local minima

00:51:54 That's a hot topic who would like to pick up

00:51:57 You should answer it art. Well, I'll answer it in one way and that is that that I think that

00:52:04 When the experts are still battling amongst themselves, there's no sense in setting up an expert system on the other hand

00:52:11 There's a lot of tedium in setting up the input for any of these programs takes a lot of

00:52:17 basic grunt work to do it and I think that

00:52:20 Expertise exists in those areas and could be encoded to make the interface for for preparation

00:52:27 Much more facile. I don't know if anybody here has any other comments about it

00:52:34 Okay, if not, Dr. Blaney, Eric is on line two

00:52:39 Hi

00:52:41 If you have a series of five or six compounds with diverse

00:52:45 and flexible chemical structures

00:52:47 That seem to bind to the same site based on biochemical evidence, but there's no obvious way to relate them

00:52:53 Are there computational techniques that can help define the karma for pharmacophore?

00:52:58 There

00:53:00 There are some but I think you're still going to have to rely a lot on

00:53:04 What you probably call chemical intuition?

00:53:07 Approaches that seem to be popular for that and will work occasionally maybe

00:53:11 Generate an electrostatic potential surface for example and try and align the molecules in that way

00:53:16 You'll find several papers in the journal molecular graphics and maybe some in journal of computational chemistry

00:53:22 from Phil Dean's group describing

00:53:24 Methods of doing that I think with with me it usually comes down to

00:53:29 trying to find

00:53:31 Which sites in each one of the molecules have structure activity relationships that are similar?

00:53:36 I might look for example for an aromatic substituent and a position on aromatic ring that has the same behavior in each different class of

00:53:44 Structures and then from infer infer from that that they might overlap the same way

00:53:49 That's always a challenging problem. We typically will try several orientations

00:53:54 Until we hopefully converge on one that we think is unique

00:53:57 Our last question is for anyone on the panel it is from from Arthur on line five

00:54:03 Hi, I'm calling from Storrs, Connecticut, and I'm interested in finding out

00:54:08 besides the

00:54:10 Distance geometry approach that has been

00:54:13 very well discussed thus far

00:54:15 There there apparently are some other approaches in general that have not been discussed in any great detail to this point

00:54:23 What I'm getting at is the burning issue. I guess amongst additional chemists and agricultural chemists is

00:54:30 an inability to be able to take a large number of compounds and create let's say a hypothetical active site or a

00:54:38 receptor site to model into

00:54:41 Would like to address this question in general to the panel and I'll await your response

00:54:45 All right

00:54:49 The the approach is available right right now

00:54:53 You know there's the confer approach that was published by Kramer and Bunce that Peter talked about I

00:54:58 Think that one looks very encouraging

00:55:00 Provided you already have an idea of your own of how to overlap these molecules in a pharmacophore

00:55:05 That's always going to be your bottleneck. Okay. You have a must interrupt. I'm very sorry

00:55:10 I want to thank our panelists David case Peter Coleman Bill Jorgensen and Jeff Blaney and our major underwriter digital equipment corporation

00:55:23 Question what computer company has the largest number of software applications for chemists answer

00:55:32 digital equipment corporation

00:55:35 question

00:55:36 What company is the leader in providing integrated solutions for research scientists answer

00:55:43 digital equipment corporation

00:55:46 question

00:55:48 What company recently added vector processing to their family of computer systems?

00:55:53 answer

00:55:55 digital equipment corporation

00:55:58 Still things are not so black and white in today's world of computational chemistry and digital knows that

00:56:06 For years, we've been working closely with scientists helping them search for the unknown and make new discoveries

00:56:12 in fact, the first vac system ever sold was purchased to perform advanced computational chemistry applications and

00:56:20 Digital is collaborating with the quantum chemistry program exchange as well as with other software suppliers that provide chemistry related software

00:56:29 Presently digital is working with Dr. Peter Coleman and associates at the University of California, San Francisco

00:56:34 To optimize amber for the backs 9,000 and back 6,000 computers with vector processing options

00:56:41 The university will make this version available to the scientific community

00:56:45 the backs 9,000 systems with vectors deliver supercomputer performance and

00:56:50 Like other backs is the backs 9,000 is fully compatible with our entire family of backs computers and workstations

00:56:57 All offer a choice of either digitals VMS or Ultrex operating systems

00:57:02 Our risk-based deck system family of high-performance workstations and systems running Ultrex

00:57:07 Demonstrates our continuing commitment to open systems, and there is more to come

00:57:13 We know how important visualization technologies are to molecular biology

00:57:18 Digital backs and risk workstations along with leading public and commercial visualization software provide a broad set of scientific

00:57:26 visualization tools our commitment in this area is strong and active

00:57:31 For presentation of research results digital's compound document architecture integrates text and graphics even from different systems

00:57:39 into scientific documents

00:57:42 Scientific collaboration and communication depend on network technology, which links the research community locally and worldwide

00:57:51 Digital is the leader in supporting network standards and in providing advanced network technology

00:57:57 Digital products are designed for use in a distributed environment

00:58:00 With systems from other vendors connecting resources including supercomputers

00:58:06 From the lab bench to the supercomputer from your desk to information sources worldwide

00:58:11 Digital has the integrated computing solutions for your research environment

00:58:17 Digital Equipment Corporation make that discovery

00:58:30 I

00:58:41 Know that you the viewers have received a great deal of information today

00:58:45 If you have more questions

00:58:47 You can contact our panelists by calling their institutions or by getting their phone numbers from the San Diego

00:58:54 supercomputer Center or the American Chemical Society

00:58:57 One last special thanks this to Dr.

00:59:00 Roseanne Steckler from the San Diego Supercomputer Center for acting as technical consultant and lining up our speakers for today

00:59:08 On behalf of the supercomputer Center, San Diego State University and the American Chemical Society

00:59:15 Continuing Education Department. I want to thank all of you for joining us. We hope you have found the program informative

00:59:22 Please don't forget to complete your course evaluations and return them to ACS and please note that the American Chemical Society's

00:59:30 Next video conference on polymer properties is scheduled for March 16th

00:59:35 1990 the ACS will be announcing their lineup of courses for the autumn very soon. We hope you'll join us again

00:59:42 Thank you and good day