Transcript slides - NMRbox

Frank Delaglio
June 21 2016
Version 3
Identify many H-H short range
NOE distances
Supplement with torsions from
J-Coupling values
Assume standard peptide
geometry
Use simulated annealing to find
a structure which matches
distances
Identify many H-H short range
NOE distances
Supplement with torsions from
J-Coupling values
Assume standard peptide
geometry
Use simulated annealing to find
a structure which matches
distances
The network of distances is
complicated, likewise the NOE
spectra are complicated
NOE distances are only
qualitative
A given peak might be the only
evidence of an interaction
A mis-assigned peak can be
similarly problematic
F
F
Y
Beta Sheet
Y
Alpha Helix
F
[email protected]
CA vs CB Chemical Shift Colored by Amino Acid Type
Subtract Residue-Specific Random Coil Shift to
form Secondary Shift
Chemical Shift and Backbone Structure Motif
Match database triplet with target, based on sum-of
squares difference in chemical shifts, plus residue
type homology term.
Use central residue as predictor of phi and psi.
Yang Shen, and Ad Bax, J. Biomol. NMR, 56, 227-241(2013).
The TALOS-N database contains 580 proteins.
On average, TALOS-N makes consistent predictions for
about 90% of the residues.
About 3.5% of the unambiguous predictions made by
TALOS-N differ from the crystal structure.
On average, the RMSD as reported by TALOS-N for the
consensus predictions was 8.7 degrees for φ, and 8.5
degrees for ψ.
The actual RMSD of the "correct" predictions relative to
the crystal structures was 12.3 degrees for φ, and 12.1
degrees for ψ (which includes the uncertainty in the X-ray
derived angles).
TALOS-N can identify the chemical shift signature of a
given χ1 rotamer for about 50% of the residues, all
corresponding to cases where no extensive rotamer
averaging is taking place.
http://spin.niddk.nih.gov/bax/nmrserver/talosn
Y Shen and A Bax: Protein backbone and sidechain torsion angles predicted
from NMR chemical shifts using artificial neural networks. J. Biomol. NMR, 56,
227-241 (2013).
SPARTA+
http://spin.niddk.nih.gov/bax/nmrserver/sparta
Y Shen and A Bax: SPARTA+: a modest improvement in empirical NMR
chemical shift prediction by means of an artificial neural network. J. Biomol.
NMR, 48, 13-22 (2010).
Consistent blind protein
structure generation from NMR
chemical shift data
Proc Natl Acad Sci USA, (2008) 105,
4685-4690
Yang Shen
Oliver Lange
Frank Delaglio
Paolo Rossi
James M. Aramini
Gaohua Liu
Alexander Eletsky
Yibing Wu
Kiran K. Singarapu
Alexander Lemak
Alexandr Ignatchenko
Cheryl H. Arrowsmith
Thomas Szyperski
Gaetano T. Montelione
David Baker
Ad Bax
Using SPARTA Chemical Shift Prediction to
Improve ROSETTA Scoring Function
0.57
0.69
0.64
0.66
0.60
2.07
0.70
1.10
2.03
Structures of Two Designed Proteins with High
Sequence Identity
NMR structures of Ga88 and Gb88
NMR structures vs csRosetta models
Patrick A. Alexander, Yanan He, Yihong Chen, John Orban,
and Philip N. Bryan
PNAS, 2007, 104:11963-11968
PNAS, 2008, 105:14412-14417
Mean-to-mean backbone RMSD
1.31A
1.07A
NMR Structure
CS-Rosetta Structure
https://csrosetta.bmrb.wisc.edu/csrosetta?key=6106ea029d9c
Luke Arbogast




Robert Brinson
Yves Aubin
John Marino
2D HN/N correlated NMR to reveal High Order Structure
Practical 2D H/C correlated Methyl NMR of mAbs at natural abundance
Multivariate analysis for easy evaluation of NMR fingerprint data
NMR spectral fingerprinting without spectra
Goal: Use NMR to provide direct answers about properties such as
protein fold, excipient effects, glycosylation, stability, and aggregation.
Changes in these properties can reduce the efficacy of a biotherapeutic,
or cause harmful immune responses.
Strategy: Use 2H, 13C, and 15N isotopic labeling to guide the development
of natural abundance methods.
[email protected]
Fourier transforms are used to convert time-domain data to frequencydomain, and the information content is similar in all domains.
Time-Domain
Interferogram
=
Frequency Domain
=
Conventional FID
NUS FID
Interferogram
NUS Interferogram
Fourier Spectrum
NUS Fourier Spectrum
Conventional FID
NUS FID
Interferogram
NUS Interferogram
Fourier Spectrum
NUS IST Reconstruction
Scaled Intensity
FT vs NUS
FT vs FT
Scaled Intensity
Scaled Intensity
Top 10 Drugs by US Sales Accounts for $60 Billion of $315 Billion
Four of the Top 10 are Biologics
Product
Sales 4/2014 - 3/2015
Used For
Type
Humira
$8,290,106,091
Anti-inflammatory
mAb
Abilify
$7,995,192,015
Antipsychotic
Small molecule
Sovaldi
$6,957,331,432
Hepatitis C
Small molecule
Crestor
$5,958,997,432
Reduce cholesterol
Small molecule
Enbrel
$5,953,627,734
Autoimmune diseases
Protein/IgG1
Harvoni
$5,398,133,616
Hepatitis C
Small molecule x 2
Nexium
$5,394,307,899
Reduce stomach acid
Small molecule
Advair Diskus
$4,789,231,826
Asthma/COPD
Small molecule x 2
Lantus Solostar
$4,770,782,304
Diabetes
Protein
Remicade
$4,614,448,608
Autoimmune diseases
mAb
Biologics are life-changing and life-saving therapeutics. They are also expensive, an issue for
everyone in the healthcare system. As originator biologics go off patent, less expensive biosimilars
can be produced. Development and manufacture requires monitoring high-order structure,
aggregation, stability, and modifications such as glycosylation.
Statistics from IMS Health Inc. as quoted on medscape.com, May 6, 2015
Most Drugs work by Targeting a Specific Protein or Class of Proteins
Most familiar drugs are small organic molecules, often discovered by synthetic chemists
making many variations of a molecular scaffold. Often, more than one kind of drug can
bind to a target, which also means, often one given drug will also bind to undesired
targets, causing side effects. This is often discovered late in the development process,
and is why many new drugs fail (failure rate is ~90% and hasn’t changed much).
Salicylic acid
Indomethacin
Aspirin
Acetaminophen
COX-2: Cyclooxygenase-2 (prostaglandin synthase-2, blue)
complexed with indomethacin (red)
Ibuprofen
Proteins Themselves Can be Used as Drugs:
Biologic Therapeutics
Insulin
Epogen (Erythropoietin)
Filgrastim (G-CSF)
Regulates
Glucose Levels
Stimulates Red Blood Cell
Production
Stimulates White Blood Cell
Production
51 amino acids
166 amino acids
177 amino acids
Hierarchy of Protein High Order Structure
If all of these aren’t right, and don’t stay right, the protein therapeutic is wrong. Changes in
these properties can reduce the efficacy of a protein-based therapeutic, or cause dangerous
immune responses.
MET
GLN
ILE
PHE
VAL
LYS
THR
LEU
THR
GLY
LYS
THR
ILE
THR
LEU
GLU
VAL
GLU
PRO
SER
Tertiary Structure
(Protein Fold)
Primary Structure
(Amino Acid Sequence)
Quaternary Structure
(Complex or Aggregate of Two or
More Proteins)
Secondary Structure
(Helix, Sheet, Turn, Coil)
Harder to Measure
Unfolded G-CSF, 15N Labeled
Native G-CSF, 15N Labeled
Y Aubin, DJ Hodgson, WB Thach, G Gingras, and S Sauvé: Monitoring Effects of Excipients, Formulation
Parameters and Mutations on the High Order Structure of Filgrastim by NMR. Pharm Res., 32, 3365-3375 (2015).
15N
ppm
Minor Oxidized
Species
1H
1H
ppm
ppm
Knowledge of NMR assignments and structure allows careful peak-by-peak analysis which can
correlate spectral changes with specific and subtle structural details such as a sidechain
reorientation. NMR assignment is complicated, and generally requires 13C / 15N labeled protein.
Y Aubin, DJ Hodgson, WB Thach, G Gingras, and S Sauvé: Monitoring Effects of Excipients, Formulation
Parameters and Mutations on the High Order Structure of Filgrastim by NMR. Pharm Res., 32, 3365-3375 (2015).
Antibody Proteins as Drugs: A Natural Source of Diverse Binding Partners
Instead of synthesizing and testing large numbers of small organic molecules, genetic
engineering can be used to select and duplicate antibodies that bind with high affinity
and specificity to most any target … humans generate about 10 billion antibody variations
...
IgG Antibody, ~150 kDa
Two identical heavy chains, two identical
light chains, symmetric.
Glycans at two amino acids
Variable Regions in blue and purple
Pharma Loves mAbs and Igs – Find Hit Quickly, Re-use Biomanufacturing Platform
Fv fusion
mAb
conjugate
mAb
mAb
mAb
mAb
mAb
mAb
mAb
Fc fusion
mAb
mAb
mAb
mAb
protein
mAb
Fc fusion
virus
peptide
mAb
mAb
mAb
mAb
mAb
mAb
mAb
mAb
mAb
mAb
mAb
Example from Amgen Drug Development Pipeline - Adapted from www.amgenpipeline.com
mAb
NIST Principal Investigators: John Schiel and Trina Formolo
 Standard Reference Material: issued under NIST trademark
with specified property values and associated uncertainties.
 Humanized mAb (IgG1κ) expressed in murine culture.
 Frozen bulk “Drug-like substance” donated by MedImmune.
 Extensive interlaboratory characterization by 65+ Biopharma,
Instrument, Academic, FDA participants.
 Data Publically available at igg.nist.gov
Amino Acid Sequencing
Amino Acid Analysis
N- and C-terminal Sequencing
Peptide Mapping by MS
S-S Bridge Analysis
Glycosylation Analysis
Molecular Weight Information
Isoelectric Focusing
SDS-PAGE
Extinction Coefficient
Post-Translational Modifications
Spectroscopic Profiles: CD, NMR
LC: SEC, RP, IEX
NIST RM 8670
NISTmAb 150 kDa, 4 Chains, Symmetric
Model based on Protein Data Bank Structures 1HZH 2GJ7 3IXT
Size of a mAb
~1,300 Amino
Acids
Amino
Acid
Count
Malate Synthase G
723 Amino Acids
Protein Data Bank (PDB) NMR Structure Depositions by Year
Since mAbs are much larger than the proteins usually studied by NMR, expectation is that
fingerprinting mAbs by NMR would not be practical, especially without isotopic enrichment.
Methyl groups are excellent reporters of protein fold, and 13C has higher natural
abundance than 15N (1.07% vs 0.33%). Rapid rotation of methyl groups mitigates
the effects of slow molecular tumbling in large proteins, for greatly improved spectra.
Non-Uniform Sampling (NUS) can further increase spectral quality obtainable in a
given amount of measurement time, making NMR fingerprinting of mAbs practical.
Uniform Sampling
50% NUS
NISTmAb
Methyl Groups
Ala Ile Leu
Met Thr Val
LW Arbogast, RG Brinson, and JP Marino: Mapping
Monoclonal Antibody Structure by 2D 13C NMR at
Natural Abundance. Anal. Chem., 87,3556–3561 (2015).
G2F Glycan: ~55%
G1F Glycan: ~30%
G0F Glycan: ~15%
Two identical heavy chains of the
Native NISTmAb each have a glycan
bonded to the sidechain N of Asn 297
b1-4 galactosidase
G0F Glycan: ~100%
PNGase F
Quenched after Partial Reaction
Deglyosylated NISTmAb: ~40%
Asparagine 297
Aspartic Acid
30 spectra shown in overlay, normalized to uniform maximum intensity, colored by sample type.
The four 16-scan spectra in the series are not shown for high noise.
Each spectrum is represented exactly
as a single object in a multdimensional
space.
The coordinates of the object are all of
the spectral intensities.
In this representation, similar spectra
cluster together.
Spectra with some features in common
lie along lines and curves.
PCA projects this space to a small
number of dimensions along directions
of maximum variance, so that it can be
readily viewed and characterized.
Component 3 Score
Component 2 Score
Component 3 Score
G0F
Native
Partial
Deglycosylation
40% NUS
Native
Component 2 Score
As shown, the Native, G0F, and Deglycosylated samples are well-clustered. Note also that the NUS
reconstructions are systematically different from the conventional data. In practice, this kind of PCA
analysis is very sensitive to processing details such as baseline correction and phasing.
Y Aubin, DJ Hodgson, WB Thach, G Gingras, and S Sauvé: Monitoring Effects of Excipients, Formulation
Parameters and Mutations on the High Order Structure of Filgrastim by NMR. Pharm Res., 32, 3365-3375 (2015).
pH 6.2
pH 5.5
pH 5.0
pH 4.5
pH 2.1
pH 4.0
pH 3.5
pH 2.6
pH 3.0
pH 3.4
The PCA analysis is
accomplished in seconds,
without the need for peak
detection or assignment.
PCA and other methods of
multivariate analysis can
reveal systematic behavior
and outliers that might be
hard to identify directly
from inspection of spectra,
even for an expert.
Multivariate approaches
become more useful with
larger numbers of spectra,
without becoming harder to
do.
Fourier transforms are used to convert time-domain data to frequencydomain, and the information content is similar in all domains.
Time-Domain
Interferogram
=
Frequency Domain
=
PCA on Spectra
Of G-CSF
PCA on Equivalent
Interferograms
of G-CSF
PCA on Spectra
Of G-CSF
PCA on Equivalent
Interferograms
of G-CSF
Apodization
Removed in the
Indirect Dimension
PCA on Spectra
Of G-CSF
PCA on Equivalent
Interferograms
of G-CSF Masked
with 50% NUS
Schedule
Apodization
Removed in the
Indirect Dimension
PCA Results on Interferograms of G-CSF 50% NUS
Component 2 vs 3
PCA Results on Interferograms of G-CSF 50% NUS
Component 2 vs 3
 2D H/C methyl spectral fingerprinting is practical at natural abundance
for molecules as large as mAbs.
 Multivariate methods such as PCA can be used on spectra to reveal
details of High Order Structure, post-translational modification, and
excipient effects.
 Since NMR fingerprinting can potentially be performed without the need
to identify peaks, it might be possible to develop even more efficient
measurement strategies which do not produce spectra that can be
analyzed visually, but nevertheless encode all the structural information
of interest.
 Labeled samples in preparation now will allow us to explore backbone
and sidechain dynamics, dipolar couplings, NMR HD exchange, etc.,
and relate these to high order structure, glycosylation, aggregation, and
stability.
 If sufficient numbers of spectra can be measured, NMR spectral
fingerprinting is a good potential target for machine learning
approaches.
NIST Disclaimer: Certain commercial equipment, instruments, and materials are identified in this presentation in order
to specify the experimental procedure. Such identification does not imply recommendation or endorsement by the
National Institute of Standards and Technology, nor does it imply that the material or equipment identified is necessarily
the best available for the purpose.