Title: Coronary artery disease most commonly due to atherosclerosis is one of the major causes of mortality
1Coronary artery disease (most commonly due to
atherosclerosis) is one of the major causes of
mortality and morbidity in the UK
2Coronary artery disease (most commonly due to
atherosclerosis) is one of the major causes of
mortality and morbidity in the UK
1 in 3 will suffer CAD at some time in their
life Over 1,000 deaths per million population
per year attributable to CAD in
2000 Treatment (medical and surgical)
increasingly effective but expensive and
difficult to target
3Coronary artery disease (most commonly due to
atherosclerosis) is one of the major causes of
mortality and morbidity in the UK
?
THE BIG QUESTION
How do we identify the people with CAD requiring
treatment at an early enough stage to effectively
treat them ?
4Invasive ANGIOGRAPHY is the current gold standard
for diagnosing CAD
An angiography cath lab, like this one, costs
1.5 million to equip, a further 0.4 million per
year to staff. The marginal cost of
each angiography performed is approximately 900.
A X-ray dye is injected into the artery to allow
it to be directly visualised by the X-ray camera
5Invasive ANGIOGRAPHY is the current gold standard
for diagnosing CAD
A severe blockage is clearly visible
LAD
Cx
An angiography cath lab, like this one, costs
1.5 million to equip, a further 0.4 million per
year to staff. The marginal cost of
each angiography performed is approximately 900.
A X-ray dye is injected into the artery to allow
it to be directly visualised by the X-ray camera
6Angiography
ADVANTAGES
DISADVANTAGES
Accurate and reproducible Provides
anatomical detail required for PTCA
Well-established
Invasive procedure Associated morbidity
and even mortality Expensive
Consequently, there is demand for a non-invasive
diagnostic procedure (such as a blood test) which
is completely safe and cheaper than
angiography. Such a test would be useful to
identify the 20-30 of patients who currently
undergo angiography only to discover that they
have normal coronary arteries
7Current markers for CAD
GENES
PROTEINS
ACE I/D PAI-1 4G/5G ApoE e2/e3/e4 LDL-R
Fibrinogen PAI-1 C-reactive Protein
METABOLITES
LDL-cholesterol HDL-cholesterol Triglycerides Gluc
ose
8These markers are well validated in
cross-sectional studies but have little
diagnostic power
LDLHDL cholesterol ratio
9Even multivariate analysis of these markers
rarely identifies more than 10-fold excess risk
10An ideal clinical application for high data
density analytical approaches
GENES
PROTEINS
Genomics - SNPs Transcriptomics
Proteomics
METABOLITES
Metabonomics - NMR - Mass spec
11A 600MHz 1H-NMR spectrum of human serum
Requires 150µl of serum prepared in the same way
as for classical biochemical tests in routine use
in hospital labs
Lipids
Lactate
Arginine
N-acetyl glycoprotein
Lipids
Choline
Sugars
Alanine
Glutamine
The dominant water signal is suppressed by
selective irradiation, and the resulting spectrum
is phased, baseline corrected and referenced
to lactate at 1.33ppm
12NMR spectroscopy of human serum is very
reproducible
A single serum sample from a healthy
volunteer was split into 14 replicate aliquots
and frozen Eight aliquots were thawed and
spectra generated on day 0 The remaining six
aliquots were thawed and analysed on day
14 The intra-day and inter-day reproducibility
of the NMR spectrum was calculated for 256
integral bins
13NMR spectroscopy of human serum is very
reproducible
14NMR spectroscopy of human serum is very
reproducible
Virtually all of the variation is time
independent The mean Cvar across the range
0.6ppm to 4.5ppm is 1.8 Within the range
where most of the information content of the
spectrum resides, the mean Cvar is 1.1 The
NMR spectrum is highly reproducible when compared
to other analytical techniques (eg. ELISA)
15TO BE USEFUL FOR CLINICAL DIAGNOSIS THE NMR
SPECTRUM MUST CONTAIN INFORMATION WHICH IS UNIQUE
TO THE INDIVIDUAL AND WHICH IS STABLE OVER TIME
16 The NMR spectrum encodes temporally stable
inter-person variation
Serum samples were prepared from 17
volunteers over a period of 3 months Samples
were taken at 0, 7, 16, 37, 67 and 97
days Spectra were prepared for each sample,
allowing variance due to temporal variation
within the same individual to be separated from
variance between individuals
17 The NMR spectrum encodes temporally stable
inter-person variation
Lipids
Sugars
Variance due to analytical imprecision has been
subtracted
18 The NMR spectrum encodes temporally stable
inter-person variation
Lipids
Sugars
Variance due to analytical imprecision and
temporal variation has been subtracted
19 The NMR spectrum encodes temporally stable
inter-person variation
There is statistically significant stable
inter-person variation in the NMR spectrum The
inter-person Cvar is 5 across the
information- dense region of the
spectrum There is considerable variation from
bin to bin in the stable inter-person variance
from almost 0 to over 15
20The variance component that is stable with time
is unlikely to results from environmental
variation such as diet Is this variation genetic
in origin? If it is, then metabonomics and
genomics might be accessing overlapping
information about the people under study
21 The stable inter-person variation in the NMR
spectrum is not primarily under genetic control
Serum samples were obtained from 150 pairs of
twins (half DZ and half MZ) All were women
aged 50-70 living in the UK Spectra were
prepared for each sample, and classical twin
modelling was performed for each NMR bin
to estimate the contribution of AGE, ADDITIVE
GENETIC, COMMON ENVIRONMENT and UNIQUE
ENVIRONMENT variance components
22 The stable inter-person variation in the NMR
spectrum is not primarily under genetic control
Sugar region
Genetic (A)
Unique environment (E)
Common environment (C)
Age (L)
ACEL model is shown for each bin, not necessarily
the statistically most parsimonious model
23 The stable inter-person variation in the NMR
spectrum is not primarily under genetic control
Little evidence for a genetic component
contributing to the variance in the vast majority
of bins Metabonomic and genomic variance are
likely to be largely orthogonal Using both
methodologies together on the same individual is
likely to have synergistic diagnostic
power Surprisingly, COMMON ENVIRONMENT seems
to be a major contributor to the stable metabolic
profile
24Clinical MetabonomicsA pilot study in CAD
Healthy
Disease
25Clinical MetabonomicsA pilot study in CAD
Healthy
Disease
26Unsupervised PCA suggests that there are patterns
associated with CAD
Healthy individuals
Severe heart disease
Plot of the second and third statistically
significant principle components from the PCA
model constructed using spectra from
40 individuals of each class
27Impact of Orthogonal Signal Correction (OSC)
Healthy individuals
PCA prior to OSC
PCA after one OSC filter
Severe heart disease
Application of OSC results in almost complete
separation of the two classes
28IDENTIFYING NEW BIOMARKERS
significant
lipids
choline
lipids
Loadings plots underscore the importance of
subtle compositional differences among
lipoproteins in the diagnosis of heart
disease, but also identify serum choline as a
novel biomarker of disease
29VALIDATION
1. Predicitivity of spectra not used
to construct the model More than 90 at 95
significance level 2. The same NMR bins are
contributing to separation before and after
application of OSC
30CONCLUSIONS
- Chemometric analysis of 600 MHz 1H-NMR spectra of
human serum samples is able to distinguish
diseased and healthy individuals - Similar techniques can classify on the basis of
disease severity as well as presence or absence - Application of OSC significantly improves
classification - Subtle variations in lipoprotein composition are
the dominant contributors to the model - Multivariate analysis of existing risk factor
data considerably less discriminating
Can such a test find diagnostic utility in a
clinical setting?
31OVERNIGHT COURIER
STEP 2
STEP 1
The technician dilutes the serum and runs an NMR
profile
Obtain a single blood sample from
patient. Prepare serum
15 MINUTES
3 HOURS
STEP 3
STEP 4
The expert system searches for patterns within
the profile
The consultant is provided with a diagnosis
15 MINUTES
Analytical laboratory
Hospital
ETHERNET
TIME TAKEN FOR DIAGNOSIS Less than 24 hours
32Here is a summary of the results for a pilot
study with approximately 100 patients Of those
with severe heart disease 6 were rejected by
the expert system 6 were diagnosed as
unclassified 88 were correctly diagnosed as
Diseased 0 were incorrectly diagnosed as
Normal Of those with normal coronary
arteries 0 were rejected by the expert
system 4 were diagnosed as unclassified 92
were correctly diagnosed as Normal 4 were
incorrectly diagnosed as Diseased
Metabolic profiling out-performs
existing diagnosis methods based on blood samples
33Metabonomic and Genomic Investigation of
Coronary Artery Disease
THE MaGiCAD STUDY http//www.med.cam.ac.uk/magicad
AIMS To evaluate clinical metabonomics for
diagnosing CAD in a clinical environment To
compare metabonomics and genomics with
existing methods of diagnosing CAD To
investigate risk-factors for CAD in the East
Anglian population
34Metabonomic and Genomic Investigation of
Coronary Artery Disease
THE MaGiCAD STUDY http//www.med.cam.ac.uk/magicad
Collect serum, plasma, urine, DNA and RNA from
more than 3,650 patients attending for coronary
angiography in a randomised, multicentre trial
design in 2 phases Almost 300 have been
collected to date
35Metabonomic and Genomic Investigation of
Coronary Artery Disease
THE MaGiCAD STUDY http//www.med.cam.ac.uk/magicad
Collect serum, plasma, urine, DNA and RNA from
more than 3,650 patients attending for coronary
angiography in a randomised, multicentre trial
design in 2 phases Almost 300 have been
collected to date Obtain metabolic and genetic
profiles
36Genomic approaches
Gene chip technology allows estimates to be made
of the expression level of almost 4,000 genes in
leukocytes obtained from the patients
Data courtesy of Dr Peter Ellis
37Genomic approaches
Gene chip technology allows estimates to be made
of the expression level of almost 4,000 genes in
leukocytes obtained from the patients
and multivariate analysis of these arrays will
be used in parallel with the metabolic spectrum,
potentially resulting in the most
accurate diagnosis.
Data courtesy of Dr Peter Ellis
38Metabonomic and Genomic Investigation of
Coronary Artery Disease
THE MaGiCAD STUDY http//www.med.cam.ac.uk/magicad
Collect serum, plasma, urine, DNA and RNA from
more than 3,650 patients attending for coronary
angiography in a randomised, multicentre trial
design in 2 phases Almost 300 have been
collected to date Obtain metabolic and genetic
profiles
39Metabonomic and Genomic Investigation of
Coronary Artery Disease
THE MaGiCAD STUDY http//www.med.cam.ac.uk/magicad
Collect serum, plasma, urine, DNA and RNA from
more than 3,650 patients attending for coronary
angiography in a randomised, multicentre trial
design in 2 phases Almost 300 have been
collected to date Obtain metabolic and genetic
profiles Accurately assess the phenotype of
the patient
40Metabonomic and Genomic Investigation of
Coronary Artery Disease
THE MaGiCAD STUDY http//www.med.cam.ac.uk/magicad
Collect serum, plasma, urine, DNA and RNA from
more than 3,650 patients attending for coronary
angiography in a randomised, multicentre trial
design in 2 phases Almost 300 have been
collected to date Obtain metabolic and genetic
profiles Accurately assess the phenotype of
the patient Compare the various diagnostic
approaches
41Metabonomic and Genomic Investigation of
Coronary Artery Disease
THE MaGiCAD STUDY http//www.med.cam.ac.uk/magicad
Collect serum, plasma, urine, DNA and RNA from
more than 3,650 patients attending for coronary
angiography in a randomised, multicentre trial
design in 2 phases Almost 300 have been
collected to date Obtain metabolic and genetic
profiles Accurately assess the phenotype of
the patient Compare the various diagnostic
approaches
RECRUITMENT COMPLETE Mid 2004
ANALYSIS COMPLETE Early 2005
42Cambridge University David Grainger David
Mosedale Jim Metcalfe John Fletcher Peter
Ellis Papworth Hospital Sarah Hayns Claire
Nugent Caryl Barnard Hester Goddard Duncan
McNab Sarah Clarke Hugh Bethell Peter Schofield
Imperial College Joanne Brindle Elaine
Holmes George Tranter Henrik Antti Jeremy
Nicholson St Thomass Hospital Kourosh
Ahmadi Tim Spector GlaxoSmithkline Elaine
McKilligin