Transcript Introduction to NMR Metabolomics

Metabolomics
Metabolome Reflects the State of the Cell, Organ or Organism

Change in the metabolome is a direct consequence of protein activity changes
•

Not necessarily true for genomic, proteomic or transcriptomic changes
Disease, environmental factors, Drugs, etc., perturbs the state of the metabolome
•
Provides a system-wide view of the organism or cell’s response
NMR Metabolomics Overview
Prepare the cells, tissue or biofluids
Harvest the metabolome
Collect the NMR data
Analyze the NMR data
Analyze the metabolite changes
NMR Metabolomics Data
One-dimensional
1H NMR spectrum
Two-dimensional
NMR spectra
2D 1H-13C HSQC Experiment – workhorse of metabolomics
Correlates all directly bonded 13C-1H pairs
generally requires 13C-labeling (1.1% natural abundance)
2D 1H-1H TOCSY Experiment – workhorse of metabolomics
Correlates all 3-bonded 1H-1H pairs in a molecules
NMR Metabolomics Process
Differential NMR Metabolomics
Monitor in vivo protein and drug activity
Inactive Drug
Active & Not Selective
Active & Selective Drug
Active Against Wrong Protein
Forgue et al. (2006) J. Proteome Res. 5(8):1916-1923
Halouska & Powers (2006) J. Mag. Res. 178:88-95
NMR and Multivariate Statistics
Extreme Sensitivity to Experimental Differences

Want PCA Clustering to Result from Metabolome Change NOT Experimental
Variability

EVERYTHING should be a CONSTANT between samples or the study is invalid
NMR
experimental
parameters
temperature
Buffer (pH)
shimming
Tuning &
matching
lock
90o pulse
acquisition
parameters
Spectral
width
Data points
Recycle
time
Acquisition
time
Solvent
removal
Receiver
gain
processing
parameter
Zero filling
Baseline
correction
Window
function
Linear
prediction
Solvent
removal
phasing
Differential NMR Metabolomics
Negative Impact of Noise in NMR PCA Clustering
Single NMR Sample with
repeat data collection
ATP-glucose
ATP
Higher PC2 dispersion
(-10 to 10) and an outlier
ATP #2
ATP #2
Remove Noise
ATP #9
ATP
ATP #9
ATP-glucose
lower PC2 dispersion (-4 to 2)
Differential NMR Metabolomics
The Role of NMR Signal-to-Noise in PCA Clustering
Increasing Number of NMR Scans (S/N)
Differential NMR Metabolomics
How to Quantify the Statistical Significance of Cluster Separations?

Analyze Metabolomic Data Using Tree Diagrams
• Calculate distances between cluster centers  distance matrix

Apply Standard Boot-Strapping Methods
• Randomize selection of cluster members to determine cluster center
• Generate 100 different distance matrices  100 different trees  consensus tree
• Bootstrap number -> how many times the consensus node appears in the set of 100 trees
Differential NMR Metabolomics
Bootstrap Number and Statistical Significance of Cluster Separations

Larger the Distance Between
Clusters More Significant
• Larger bootstrap
or smaller p-value
• > 50% is significant

More Data Points Easier to
Distinguish Between Clusters
• more data points (solid line)
Sample Replicates Affects Class Distinction
6
8
10
Increasing number
of replicates

Significant increase in statistical significance of cluster from a modest increase in
number or replicates
Ellipses and Tree Diagrams Define Classes


P-value on each node identifies statistical significance (< 0.001) of cluster
Ellipses represent 95% confidence limits from a normal distribution
Differential NMR Metabolomics
Metabolite Identification
S-plots
loadings
 Orthogonal partial least squares discriminant analysis (OPLS-DA)
• a non-linear variant of PCA that minimizes class (group) variations
• S-plots and loadings identify which “bins” (NMR chemical shifts – metabolites) are
strongly correlated with class separation
Differential NMR Metabolomics
Metabolite Identification
Grow cells in the presence
of a 13C-labeled metabolite
Only observe metabolites
derived from the 13C-labeled
metabolite provided to the cells
Overlay of 2D 1H-13C HSQC spectra for wildtype (red) and aconitase mutant (black)
Convert Peak Intensities to
Concentrations (HSQC0)
Our 2D 1H-13C HSQC calibration curve
Hu et al. (2011) J. Am. Chem. Soc. 133:1662-1665
Convert Peak Intensities to
Concentrations (HSQC0)
Can now compare changes between metabolites
140
Concntration (uM)
120
100
80
GPM267
GPM267DCS
GPM292
60
40
20
GPM292DCS
GPM385
GPM385DCS
MC2
MC2DCS
0
TAM23
TAM23DCS
Convert Concentrations to Heatmap
 Provides two-levels of hierarchal clustering
• Identifies replicates with same overall changes
• Identifies metabolites with correlated changes
between replicates


Provides a simple view of a large amount of data
Calculated with a statistical package, like R
http://www.r-project.org/
Differential NMR Metabolomics
Metabolite Network Mapping (Cytoscape)
Metabolites increased (red), decreased
(green) or unperturbed/undetected (grey)
Differential NMR Metabolomics
Traditional Metabolic Pathway
Some Final thoughts
 A number of different analytical
methods can be used to analyze the metabolome
• NMR, GC-MS, LC-MS, CE-MS, FTIR, etc.

A variety of statistical techniques can be used to analyze metabolomics data

Can combine multiple datasets (NMR and MS) for multivariate statistical analysis

• PCA, PLS, OPLS-DA, HCM, SOM, SVN, etc.
Can incorporate proteomics, genomics and any other data source with
metabolomics data to generate system-wide view of the organism or cell response