Gordon Research Conference on Computational Aspects of Biomolecular NMR
June 6 -11, 1999
Conference Center "Il Ciocco", Barga, Italy

G. Wagner Comment - Can detect dimer unambiguously by 15N-H1D <-> 14N-H H mixtures

                    ONLY see for dimer

June 6, 1999

Jeffrey Hoch              "Opening Remarks"

          NO NOTES

David Case                 "Where do we go from here?  Some hard problems in determining biomolecular structures and dynamics"

          NO NOTES

Lewis Kay                  "Experimental NMR Spectroscopy: A Brief Survey of an Evolving Field"

          NO NOTES

June 7, 1999

Alan Stern                  "Modern Spectrum Analysis in Multidimensional NMR: A Critical Reassessment"

     Non-linear samples    LP and maxEN

     maxEN w/ expon. sampling in two dimensions provided better S/N than LP/DFT or DFT alone.         S/N Issue

    Errors: false peaks, undetected peaks   *see   Rowland NMR Toolkit - for generating simulated NMR spectra

    Nice summary of advantages and disadvantages of LP and MaxEN

Timothy Havel          "A matrix least-squares approach to estimating distance constraints from NOESY spectra"

     Good project for David Snyder:  Havel et al. (1994) PNAS 91: 7962 (could not read easily)  2D spectra can be factorized into 3 spectra

     Can pull at J(H-H)

     getting many J(H-H) //  Yang & Havel (1994) J. Biomol. NMR 4: 807; Yang et al. J. Biomol. NMR 4: 827

     Also compared at end with Autopsy

     NOESY- peak matrix generated from COSY (i.e. could be done from chemical shift peak list) and then extract out the peak amplitudes.

     The formulation yields all intensities and does not require you to assign peaks themselves

     Issues: Accurate peak shape model
          Accurate chemical shifts
          Need also correct linewidths and multiplicities (which he gets from COSY)

     Using "noise flattening" algorithm from AUTOPSY program - and "eigen filtered"

              I think this could work much better if you (1) use 13C-edited NOESY  (2) pick high - determine structure-use structure to pick lower
              He is getting very nice peak picked NOESY spectra

Werner Braun             "Combined automated assignment and 3D structure calculations of proteins: Does it work in practice?"

          157 manually assigned constraints in initial structure
          using NO manual assignments - also get correct fold - but convergence is poor (1.5 Å)

     CsE - vs from Ram Krish - homology NOESY/TOCSY
              used coupling constants as constraints
              No manual assignments used as input

     Made comparison between manual and automatically determined constraint lists

     Major issue is peak picking - some significant differences between manual / automated structure analysis

     Using homologous structure to drive automated NOESY analysis

     Get correct fold - but not outstanding results helix tilt is pretty far off

Gaetano Montelione             "Automated Analysis of Protein NMR Spectra: Prognosis of Structural Genomics"

          NO NOTES

Frank Delaglio                     "Things a Nice Italian Boy Can Do With Couplings, Shifts, and Molecular Dynamics"

            Torsion                                    13C db for x-ray structures of resolution < 2.2 Å
            Obtained from                                    cross-correlation - 2910 predicted
            Shifts                                                               1920 correct
                                                             problem ->          58 incorrect

Bruce Johnson                     "NMRView: New tools for assignments and NOE analysis"

          NO NOTES

David Wishart                         "NMR without NOEs: Getting the most from your Chemical Shifts"

         Important that we fish out all assignments and get to bmrdb

         - NMR meets Bioinformatics
         - can use Chemical Shifts like sequence data

           380 sets of assignments in BMRDB
         1100 distinct structures
         Assignment of Protein A  ---------> Assignment of Protein B
         ORB         Gronwald et al. (1997) JBNMR 10: 165          Transferring
         GARAMM                                     JBNMR 7: 207           homologous
         SHIFTY     Wishart et al. (1997)    JBNMR 10: 329         assignments

         SHIFTY - like what we were trained to do with Norma ~1990  (access from BMRDB)

         correlation coefficient of > 90% when 90 sequence identity > 35%

         - using HNCO/HN(CA)CO  get ~90 - 100% ASSIGNMENTS

                  < 100 residues  TOCSY                                                                           | single
                      75 - 150 residues TOCSY - HSQC                                                      | experiments
                      75 - 200 residues CBCA(CO)NH                                                        | to provide
                          > 200 residues  HNCO / HN(CA)CO                                              | assignments

         should be           |             SHIFTX - computer 1H chemical shifts from 3D structure
         running this       |                       15N referencing big problem in db
         routinely on      |                  but can get r's of ~ 0.92 - good way to validate
                                                            referencing r
         all assignments/|                  data referencing 13Ca, 13C'abo good a
         3D structures  |                 Ca shift related to disulfide conformation

         Discuss with Hunter:
               He is using very simple spectra + d predicted from crystal structures to complete assignments for case where crystal structure is known.

              Also- using homology model to predict d and then use these together with minimal experimental data to complete assignments.

         Trying to use SIMPRED to predict assignments and then combine with simple TOCSY or TR experiments.

         New Protocol:
                  1) Measure d
                  2) Initial secondary structure from d's
                  3) Fold using GA + threading
                  4) Energy minimization

         Wants to apply to make assignments in bound vs. free ligand binding

         * need to get someone working on this idea that we have already discussed alot

Angela Gronenborn                     "Structures of Protein-Protein and Protein-DNA Complexes"

        could use conformational db in initial fold calculation - then eliminate in subsequent refinement

         - record as much data as we can lay our hands on
         - 5 months of measure time for 40 hD complex - complex too weak to crystallize


June 8, 1999

Rafael Brüschweiler                         "Anisotropic Motions Studied by NMR relaxation, MD, and DFT"

         It is impossible to distinguish overall motion from internal motions - this is a consequence of the inability to distinguish multiple motions  with different amplitude that occur on the same timescale - Jardetsky

                model-free                  | MD                                   | analytical
                approach                    | computer simulation         | motional
                spectral density         | Normal model                    | models
                mapping                     | analysis                              |

         MD - simulations guided model building - best we can do - Recent work from Ernst lab

         Long MD simulation statistical analysis

         analytical parameterization  compute GAF for each peptide plane of model          fitting of parameter          could ask him for their data for Mike's project

         Explained problems is setting CSA tensors from ssNMR, and liquid state NMR - proposed using quantum mechanical calculations (DFT - Density Functional Theory) to get these CSA tensors.
              -very good agreement between calculation and ssNMR values in amino acid models

         CSA tensions are changing (rapidly) during the MD trajectory - we usually consider these to be static but they are actually fluctuating

        3J(15N-13C') coupling (~1 - 2 Hz) have been measured in ubiquitin (S. Grzesiek published recently in JACS)

Gerhard Wagner                     "NMR studies of protein interactions"

         Nice example of using mutation to enhance properties of proteins - stability of aggregation

         CD2: carbohydrate essential for stabilizing protein structure (maybe true in other cases) - but can make mutant form which is stable without   sugar - this is published already Wyss, Endo
          Should look for sugar binding sites in parl - look for consensus carbohydrate binding sites

         CD58: homology model should surface exposed loops with hydrophobic groups - mutated 6 of these to hydrophilic - now get great spectra - just looked for surface hydrophobics.

Ad Bax                                     "NMR of weakly aligned macromolecules"

         using 3JHC to identify H-bond pairs!!!!  Discuss with Dani

         Very nice experiment: DNH, DNC, DC'H all measured simultaneously in 30 kD protein - provides orientation of peptide planes

          Q=[(S(Dcalc - Dobs)2)/S(Dobs)2]1/2

        can be used as a cross validation

         For bb- see good agreement between calculation (from x-ray) and observed D's.  But for side chains, not a good agreement - this has been related to dynamics.  Motional average causes dipolar coupling to be smaller than true value - used in XPLOR as "lower bound"

         Aligning a protein using purple membrane fragments to measure dipolar couplings - bacteriorhodopsin (BR)

         Bicelle orientation tensors different from BR orientation tensor - get intersections of cones

         use 591 NOEs, 945 Dipolar couplings, 90 torsional constraints provides good quality structure - but make difficult to get convergence

         Measured N-H bond length from NMR data

        GNH =1.041 ± 0.006 Å

James Prestegard                 "An order matrix approach to the structural and dynamic analysis of residual dipolar couplings in biomolecules"

     Order Matrix

     adding CTAB to brielle mixture
      DMPC:CTAB    19:1

David Fushman                     "Chromatic NMR relaxation approach to structure and dynamics of proteins"

    Field dependent relaxation measurements

    Multiple field (3 fields) => parameter of spectral density functions

     360, 520, 500 data for ubiquitin

     CSA - 120 - 200 ppm   surprisingly wide
     q = 6 - 26 °   similar measurements have been made for otherproteins

     See Fuschman et al (1998) JACS 120: 10947   for a discussion of how CSA uncertainties propagate to dynamic parameters.  Also  (1999) J. B. NMR 13: 139

Sandor Szalma                     "Methodologies for NMR Structure Refinement: Direct Refinement Using NOE Intensities, Residual Dipolar Couplings, Cross-correlated Relaxation Rates and Local Ensemble Averaging"

     see Haenggi & Braun (1994) FEBS Letts. 344:147  Removal of violations

     Have implemented cross-correlation relaxation constraints (G) and residual dipolar coupling measurements in XPLOR.  Also, implemented ensemble averages of 3J, G and NOEs.

Michael Nilges                 "Optimal use of NOE restraints"

     Ambiguities handled differently from other applications

                   recently - insulin hexamer. - many ambiguous constraints requiresmany hours to run             -use very wide bounds, "noise" exclusions
              -ensemble averages
              -trim averaging
                  vs narrow bounds
                vs Narrow bound with noise exclusion - best accuracy

     *Send Nilges CONGEN papers (*loose bounds  *shape of potential function) he is using Donna's potential, should point this out

     This idea previously done by Constantine et al.: Use violation analysis to identify inconsistent constraints - identify violated constraints - ensemble average in this region of molecular (e.g. Constantine et al.) - but ensemble averaging is done with cross validation.

     Run MD unconstrained compute <dij>, compute sij from autocorrelation of tdisx vs dis from true NOE data
  calibration with min. at distances calibrated with NOE

     deposit NOESY peaklists, not upper bounds

     used MD simulations to help in interpreting of  NOE data????  did not understand

Peter Güntert            "Torsion angle dynamics in automated NMR structure calculations"

 DYANA - torsion angle dynamics

 CAR / DYANA - introduce ambiguous constraints, introduce simulated annealing

 For each peak - discard assignments contributing £ Pvol % to peak volume

 Recalibrate distances assuming multiple contributes to each peak

 Each cycle starts from original unassigned peak list

 Robust with artifactual peaks

 apo CopZ  Cu-binding protein Wimmer et al. in press

 manual vs. automated analysis | only running 6 cycles

 * shows bundles at each of 6 stages - happy that fold is defined already at 1st cycle

 * shows cf to manual structure at each stage

        only ~1% of difference is
        assignments between manual
        and automated

        Wrong constraints:
        - Imperfect peak picking

         -artifactual peaks
         -imprecise position
         -wrong columns
Methods to improve reliability:
  * use of manual assignments
  * criteria for reliable peaks
  * multiple assignments

 Using "contact" frequency to define peak assignment reliability. Mutually supporting NOE assignments

 Ambiguous Distance Restraints: The presence of wrong assignment possibilities have little influence on the structures so long as the correct one is on the structure

 Idea: From two peaks end (e.g. selected randomly) combine assignments into one extended multiple restraints
  eg.1.  N peaks - 10% wrong, only 1% contribution???
  eg. 2  P14a  compare of CAR, X-ray, manual about as good/bad as our results

 No H-bond
 J's input, constraints but dihedral constraints computed automatically  Used manual assignment peak list - clear peak list.  But another protein - Killer toxin- using Autopsy peak list.

W. Braun - need to define reliability index for quality of assignments.

June 9, 1999

Wilfred van Gunsteren                 "The effect of motional averaging on the calculation of NMR-derived structural properties"

 *  MD simulation of b heptapeptide - ensemble average satisfies NOE and J constraints better than any single NMR structure (see Sept issue of Proteins)

 using 50 ns MD simulation at 340° K
   J. Mol. Biol. (1998) 280: 925 - 932

 75 % unfolded -> no NOE violation larger than 1 Å on average

 J values not good indications of folded / unfolded equilibrium

 NMR data has only limited capacity to distinguish - single fold from mixture A folded/unfolded conformers

 Peter Wright - also feels that J's are generally not good indicators of folded/unfolded - except when you see extreme J's (e.g. ~3 Hz)

David Pearlman                     "Improved Methods for NMR Structure Determination from NOE-derived Distances and Coupling Constants"
 Nice idea (shows advantages over Torda - type time average):  FINGAR - Fit NMR using Genetic Algorithm
  - define conform database (use MD and / or PG + cluster)
  - to each conf assign weight
  - use genetic algorithm to determine weights
 Extended FINGAR: finds bad restraints.  FINGARW does a good job of tagging bad restraints - but I did not understand this???

 Also showed example of discovering bad restraints

 Argues that FINGAR is difficult for proteins because hard to generate basis set.  Maybe can generate rotomer states in regions of protein with residual violations.  - Could compute combinatorial number of rotomer states.

Michael Levitt                "Accurate Molecular Dynamics Simulation for NMR"
 1 ns / day for MD ?
 coordinate RMS
 distance RMS
Lucia Banci                     "Computational Aspects in NMR Spectroscopy of Paramagnetic Metalloproteins"
Kathleen Hall                 "Unrestrained stochastic dynamics simulations of the RNA UUCG tetraloop using an implicit solvation model"
Joseph Puglisi                     "Solving biologically relevant RNA structures by NMR"
James Williamson             "Approaches to NMR of Large RNAs"
Juli Feigon                     "Cation Binding, folding and interactions in DNA and RNA"
June 10, 1999

Peter Wright                 "Structural and dynamical characterization of disordered states of proteins"

 Unfolded proteins have functional rules

 Kinase-induced activation domain (pKID) of CREB.

 - 1H - 15N HSQC looks like unfolded protein - but see significant spectral changes upon binding to target protein

- Should review these papers: pKID : KIX complex
     60 AA            peptide fragments at 180 residue proteins
                            80 residue fragments fully functional
                             not folded - because of truncation of C- terminal helix

    active/not folded
                                                                                        active and folded

       maybe dmPI is cut wrong

        -make it longer
        Poplar pastocyanine
  - folded at high salt (1M NaCl)
  - unfolded at low salt
     *  limited resonance dispersion

     * tendency to aggregate (need to work at 100 - 300 mM)

     * variable linewidths

     * ensemble averaged NMR parameters

     NMR Parameters in Unstructured Proteins

      most important - chemical shifts  | coupling constants not very useful
      H/D exchange
      15N relaxation

    Discussion with Hunter - use CO-TOCSY, CA-TOCSY, etc to do complete assignments

Christian Griesinger             "Computational Aspects of novel NMR parameters for the determination of structure and dynamics of biomolecules"

     Peptide - C20W - calamodulin complex

     Yb (not the best - distortion? would be better. Dp?; will give both magnetic susceptibility and pseudoconstant shifts) - DTPA (buy an anhydric and couple to N-term of peptide) - Peptide - (this peptide is then bound to calamodulin)

     gets residual dipolar coupling and pseudoconstant shifts

     gets relative orientation of two domains (using order matrix approach for using dipolar coupling data)

     Use of dipolar coupling as intervector projection constraints - take relative orientation of N-H bind vectors - form allowed / disallowed 0 values.


      using easily accessible structural info for structural bioinformatics

     Can thus use restraints as local restraints right from start - do not need ? orientation

     combine 13C C5I and dipolar couplings - use as constraints to try to map protein to structures in the PDB  (!!! locations and relative orientations of helices)

     Idea:  Can the dipolar coupling in Z-domain sort out of helix-1 distortion - can also consider to use chemical shift changes in orientation as structural constraints

Christina Redfield                 "Structure and Dynamics Probed by 15N Relaxation and Partial Sample Orientation"

     Old data - the average errors for T1, T2 were ~1.5%



     1) Never assume isotropic tumbling
     2)Need to account for a ~20°

     Need to take into account effort of non-colinearity:
          at 500 MHz - less important  efforts are more pronounced at higher
          at 750 MHz ~2%   magnetic fields and with more
          at 900 MHz ~4%   anisotyping

     Put into bundles:         5% DMPC:DHPC                     2.9:1.0          also with CTAB -
                                         7.5% DMPC:DHPC                  2.9:1.0          charge - minimize
                                         7.8% DMPC:DHPC:CMB        2.9:1.0:0.1

     with CTAB get orientation ? that matches. inertia moments, without CTAB set tilt due to electrostatic effects.  Having two different allignments useful to fit CSA ?.

     DJ=-174.4°, a=18° for axially ? CSA ? - also did analysis for asymmetric CSA tensor - similar result
              using s=1, rNA=1.04, DJ=-174.4, hasy=0.14, a=18.6°
                       s=0.94, rNH=1.02, DG=-172.4, h=0.14, a=18.7°


Hartmut Oschkinat             "Steps towards automating the structure determination process and concepts for membrane protein structures"

      NO NOTES

Deborah Kallick                "Analysis of Proton Chemical Shifts in Turns"

      NO NOTES

John Markley                    "News from NMRFAM and BMRB: Experiment Management System for Biomolecular NMR and Software Tools for Archiving and Retrieving NMR Depositions"

     * Genomics / Proteomics
     * Role of NMR in Structural / Functional Proteomics

     Experiment Management System for NMR
         Lab organization - expression vectors
          Data collection
          Data processing / analysis

     Comprehensive definity tags for NMR_Star is in submission forms for BioMagResDatabase

     May be as many as 400 new structures in 1999.
          Projections:     2000         503 structures
                                  2001         686 structures

     Future BMRB software
      * Make use of XML as alternative format to NMT-Star
      * Develop CORBA interfaces to the databases

     SeSene - NMRFam experiment management system.  Experiment management system for NMR

     Camel (have student look at this)- available from BioMagResDatabase Server (under software) Camel Client / Alibaba Server
          can archive experimental protocols (e.g. 3D experiment) from spectrometer

     Camel has nice features for scheduling instrument time

     Camel provides interface for archiving expression plasmids, cleavage sites, etc.

     Longer range projects:
          peak picks
          automatic analysis software

Robert Kaptein             "Validation of Biomolecular NMR Structures"

     Procheck NMR                  Tools for validation of structures. ~97 structures
     Aqua                                   available with constraints in 1996.  Analysis done for
     What If                               these 97.  Now ~1200


     Violating - these can be manipulated (by editing)

     rms viol. - should be less than ~0.1Å

     Many NMR structures do not have good Ramachandran plots.

     Geometric features:
          -bond lengths - OK - problems mainly due to access S-H
          -bond angles - OK
          -Ramachandran plots - bad constraints - problems in XPLOR - due to scaling   vdw radius by 0.8
          -planarity - peptide    XPLOR, DIANA - too restrictive
                  side chain planarity     XPLOR is problematic.  too loose
                                  deviations also seen in DIANA

     Nomenclature - Prochiral methylene and methyl groups often supplied.

     Completeness = observed NOEs  x 100 %
          expected NOEs

     What is "expected" NOE. HN, Ha, Hb2, Hb3, then 1 H-atom per methylene pair

     can expect to see ~60% completeness even in disordered regions - defined  specifically for each amino acid type

     -deposit constraints
     -eliminate "redundant" constraints
     -do no average coordinate
     -report conditions for all protons consistent with protination state
     -proper balance between experimental and energetic constraints


    Arria:  Provides interface to NMRView, Xeasy, Aurelia, CNS using python in the interface to CNS 30 min/structure on R10000 processor
    PECKAY: Protein Structure Elucidation by Iteration Calculation using Ambiguous  NOEs.  NOESY spectra -> XPLOR input file. Iterative constraints
    Rescue:  available  http://www.infobiogud.univ-montpl.fr    Neural networks for spin system typing
    Domaille: ? repeat Nature ~Oct 1997 p999 NMR
    Nature ~Fall 1998        Mol. Replacement in Complex
   wuld provide all data - NOESY folds different