Roy Goodacre and friends - Sciencesconf.org · Zahra Rattray, Gindo Tampubolon, Bram Vanhoutte,...
Transcript of Roy Goodacre and friends - Sciencesconf.org · Zahra Rattray, Gindo Tampubolon, Bram Vanhoutte,...
Metabolomics by numbers:
lessons from large-scale phenotyping
@RoyGoodacre
@LivUniCMR
@Metabolomics www.biospec.net
Roy Goodacre and friends
Human Metabolism
Metabolite intermediate
of metabolism
“Traditional” linear view
of a metabolic pathway
From metabolites to metabolomics
Metabolomics defined as the metabolic
complement (metabolite
pool) of a cell or tissue
type under a given set of
conditions
A “scale-free”
metabolic network
Metabolomics & biological systems
pathways networks
S
metabolites
metabolomics
proteins/mRNA
Integrate: SNP / genotype
-----> system understanding
Biofluids (exo-metabolome)
Endo-metabolome
Biopsy
Sputum
Volatilome
1y cell culture
cells footprint
www.husermet.org
Funded by and in collaboration with:
Department of Trade
and Industry
GC-MS + RPLC-MS METABOLIC
PATHWAYS
Glycolysis, TCA cycle
Pentose Phosphate
Amino acid metabolism
Gluconeogenesis
Urea cycle
Inositol metabolism
Carbohydrate metabolism
PROVIDES GOOD METABOLITE COVERAGE IN COMPLEMENTARY PATHWAYS
METABOLIC
PATHWAYS
Lipid and fatty acid
metabolism
Secondary metabolite
synthesis
Metabolism of co-
factors and vitamins
Metabolism of
Xenobiotics
LESSON I Mass Spectrometers & Chromatography DRIFT
QCs allow signal
assessment
QC 1
QC 2
QC 3
QC 4
QC 5
QC 6
QC 7
QC 8
QC 9
QC 10
blank
column test solution
sample 1
sample 2
sample 3
sample 4
sample 5
QC 11
sample 6
sample 7
sample 8
sample 9
sample 10
QC 12
sample 11
Begley P. et al. (2009) Analytical Chemistry 71, 7038-7046
Real:
60 serum (healthy).
QA:
sigma serum.
Spike1:
glutaric acid, citric
acid, alanine, glycine,
leucine,
phenylalanine, and
tryptophan. [each at 0.16 mg mL-1]
Spike 2:
caffeine & nicotine. [each at 0.16 mg mL-1]
LOESS: low-order nonlinear locally estimated smoothing function
QCs allow signal correction
Sample
QC
Instrument annual
maintenance
Before:
After:
Dunn W. et al. (2011) Nature Protocols 6, 1060-1083
QC 1
QC 2
QC 3
QC 4
QC 5
QC 6
QC 7
QC 8
QC 9
QC 10
blank
column test solution
sample 1
sample 2
sample 3
sample 4
sample 5
QC 11
sample 6
sample 7
sample 8
sample 9
sample 10
QC 12
sample 11
QC guidelines
Broadhurst, D. et al. (2018)
Guidelines and considerations for
the use of system suitability and
quality control samples in mass
spectrometry assays applied in
untargeted clinical metabolomic
studies. Metabolomics 14: 72
Q: Did the QC correction work?
Discriminant
analysis:
Attempting to
separate the
10 batches
A: Yes!
Admin
+User
+Experiment
Growth
+Treatment
+Environment
SampleHandling
+Explant
+Sample
Collection
+Collection
+Event
SamplePreparation
+Aliquot
+AnalysisMaterial
AnalysisSpecificSamplePreparation
+PreparationMethod
+Procedure
InstrumentalAnalysis
+Machine
+Run
MetabolomeEstimate
+Output
+DataPoint
BiologicalSource
+Genotype
+Source
HUSERMET pipeline
metadata capture
metabolomics curation
reference map of normal
biomarker
discovery
SOPs
QCs
D
SOP
SOP
N=1000s
Multiple samples -
Biobank for future
studies
Admin
+User
+Experiment
Growth
+Treatment
+Environment
SampleHandling
+Explant
+Sample
Collection
+Collection
+Event
SamplePreparation
+Aliquot
+AnalysisMaterial
AnalysisSpecificSamplePreparation
+PreparationMethod
+Procedure
InstrumentalAnalysis
+Machine
+Run
MetabolomeEstimate
+Output
+DataPoint
BiologicalSource
+Genotype
+Source
reference map of normal
biomarker
discovery
HUSERMET pipeline
and data analysis
Brown, M. et al. (2005) Metabolomics 1, 39-51
Mamas, M. et al. (2011) Arch. Toxicology 85, 5-17
Goodacre, R. et al. (2007) Metabolomics 3, 231-241
• Normal populations: • Stockport PCT
• GSK, EMAS
• 1200 subjects: • Took 18 months
Clinical Chemistry versus LC-MS
Correlation analysis based on 1,200 individuals
Clinical Chemistry versus GC-MS
microbial drugs food
Many molecular phenotypes
Clinical characteristics of the
Husermet cohort
Dunn, W.B. et al. (2015) Metabolomics 11, 9-26.
med
ian
(IQR
)
n
Data Analysis
RFsorSVM
PLS-DA
MetabolomicsChemometrics
Modeler
Feature
discovery
Combination of:
Multivariate
PLS, RFs, SVMs committee voting
Bootstrapping (n=1000) + permutation testing
Univariate
ANOVA + Q-Q plots (normality)
FDR: Benjamini–Hochberg procedure
Common
Features
Ageing: glycolysis and TCA
Figure: Levine A.J. & Puzio-Kuter, A.M. (2010)
Science 330, 1340-1344
2-way ANOVA; age & gender:
F(1,779)=79.8, p=3.1x10-18
Gender effects
Seen previously 4-hydroxyphenyllactic acid,
creatinine, citrate, urate, glycerol, hexadecenoic acid
Higher in Females Caffeine: food consumption
2-aminomalonic acid: associated with atherosclerotic plaques
glycerol, + glyceric acid, glycerol-3P: glycerolipid and glycerophospholipid synthesis
Marker of oxidative stress; oxidation product of methionine
2-way ANOVA on gender: F(1,901)=20.3, p=7.7x10-6
Gender and age effects
2-way ANOVA
Age (<50 vs. >64 y): F(1,788)=39.1 F(1,778)=11.7
p=6.8x10-10 p=0.0007
Gender F(1,788)=55.4
p=2.6x10-13
Q: How fast did we get to this point?
A: A lot slower than expected!!!
http://searchengineland.com/figz/wp-content/seloads/2014/08/speed-slow-snails-ss-1920-800x450.jpg
LESSON II Need to control experimental design
Factors affecting the human metabolome
body composition
tissue turnover
metabolic rate (at rest)
age
human genotype
health status
reproductive status
diurnal cycle
nutrients
non-nutrients
drugs
physical activity
microbiome
mental status
Intrinsic factors
Extrinsic factors
Metabolic status
Goodacre, R. (2007) J. Nutrition 137, 259S-266S.
Metabolomics of a
superorganism
Complex!
Its metabolites include:
Human derived metabolites
Microbial derived one
Nutritional metabolites
Xenometabolites
To do it properly…
Need to control diet
Sampling time (diurnal rhythm)
Need to have matched controls
Many molecular phenotypes
Clinical characteristics of the
Husermet cohort
Dunn, W.B. et al. (2015) Metabolomics 11, 9-26.
med
ian
(IQR
)
n
Q: Is size important?
We measured
1200 subjects
100 bootstraps for
each sample size
selection
Recommend:
at least 300
subjects per class
A: You Bet!
Dunn, W.B. et al. (2015) Metabolomics 11, 9-26.
Trivedi, Hollywood, Goodacre. New Horizons in Translational Medicine 2017: 3, 294-305
Metabolomics and study size
Underpowered many < 50 people in total
Poor Balance
Only 3 studies > 300 Case vs. 300 Control
LESSON III Biological inference is possible
fRaill: Frailty, Resilence And
Inequality in Later Life
Nicholas J W Rattray, Drupad K Trivedi, Yun Xu, Tarani Chandola,
Robert J A H Eendebak, Caroline H Johnson, Alan D Marshall, Kris Mekli,
Zahra Rattray, Gindo Tampubolon, Bram Vanhoutte, Iain R White,
Frederick C W Wu, Neil Pendleton, James Nazroo & Royston Goodacre
Unpublished data
ELSA, Fraility and Rockwood Index
• Longitudinal data from >50 year olds.
• Objective and subjective data relating to
health and disability, biological markers of
disease, economic circumstance, social
participation, networks and well-being.
• Annual 2 h questionnaire (100 Qs) over the
past 15 y following 11,400 people at Wave 1.
Cumulative Frailty Score
No.
of
Subje
cts
0 – 0.1 = 320
0.1 – 0.2 = 571
0.2 – 0.3 = 206
0.3 – 0.4 = 59
Over 0.4 = 36
2006 Rockwood papers in CMAJ and The Gerontologist
Currently there is no single generally accepted clinical definition of frailty.
Metabolomics – wave 4 (n = 1192)
Frail
Non-Frail
Pre-Frail
+
= 52-54%
= 22-28%
= 16-26%
Group Variance
Frailty Index 0 1
Fre
qu
en
cy
Bin Sample Size
0.0 – 0.1 = 320
0.1 – 0.2 = 571
0.2 – 0.3 = 206
0.3 – 0.4 = 95
Discriminant Function 1
Dis
cri
min
an
t F
un
cti
on
2
Mummichog within XCMS
mummichog:
accurate mass of significant metabolites pathway analysis
Huan, T. et al. (2017) Nature Methods 14, 461-426
Mummichog network enrichment
Findings
Several pathways
highlighted
MSI Level 1
LC-MS-MS on Orbitrap
+ Standards
Validation
Targeted analysis in
Wave 6:
50 Frail vs. 50 Resilient
Further 600 individuals
thanks to Reviewer…
Carnitines
Tocotrienols
Tryptophan
degradation
Links to ageing
Antioxidant properties of
Vitamin E analogues
Protect fatty acid groups from
lipid peroxidation.
With more free fatty acids
Carnitine shuttle generates
more acetyl-Co, vital for
electron transport chain.
Within the frail-metabotype:
Lower abundance of Vitamin
E analogues and carnitines
Indicates a down-regulation of
this system and hence a lower
energy output.
Mebo: SWOT analysis
Strengths
Emerging diverse field
Great excitement in the area
Integration with other ’omics
Weaknesses
Lack of Metabolite Id.
Semi-quantitative at best
Poor statistics/validation
Opportunities
Improving Met Id.
Dynamic measurements
Spatial metabolomics
Improved patient care
Biological understanding
Threats
Over cooking results
Working in isolation
Technology expensive “Statistics are like bikinis. What they reveal is suggestive, but what they conceal is vital”
Aaron Levenstein
Poste, G. (2011) Bring on the biomarkers. Nature 469, 156-157.
Winder, C.L. (2011) Trends Microbiol. 19, 315-322. http://metaspace2020.eu.
www.biospec.net #ScienceIsGlobal