1. Perspectives of metabolomics towards personalized medicine
Oliver Fiehn
Genome Center, University of California, Davis
fiehnlab.ucdavis.edu
PI Prof Carsten Denkert, Charite, Berlin

2. Metabolism is the endpoint of non-linear cellular regulation
Genotype x Environment
mRNA expression
Background
protein expression
metabolite levels & fluxes
temporal x spatial resolution
phenotype
Fiehn 2001 Comp. Funct. Genomics 2: 155

3. Metabolic phenotypes reflect multiple origins
SNPs allelic variants gender racial disparities inherited methylations
gut microbes
calorie intake food composition life style / exercise disease history
Background
transport

4. gate to personalized medicine
Metabotypes:
gate to personalized medicine
“Intervention”
Metabotype intensity
Disease
Background
Healthy
Time
Metabotype = personal sum of metabolic data, e.g. biomarker panel.
Analyzed over time or in response to treatment
vd Greef et al Curr. Opin.Chem.Biol. 8: 559

5. Case study of a Finnish girl diagnosed with type 1 diabetes at age 9y
Glutamate 13-fold increase
g-aminobuyrate (GABA ) 9-fold increase
g-aminobutyrate (GABA)
Glutamate
Glutamate decarboxylase antibody (GADA)
GADA
IAA
Insulin autoantibody (IAA)
Background
% max.
Diagnosis
Normal level
(GABA, Glu) + + 1 2 3 4 5 6 7 8 9
Age (years)
BCAA++, ketoleucine before GADA, IAA
Orešič et al J. Exp. Med. 205: 2975

6. Oral Glucose Tolerance Test
Challenge tests tell more if clinical chemistry is advanced to metabolomics
Oral Glucose Tolerance Test 20 40 60 80 AU
120
min
individual subjects
free palmitic acid
Background

7. But cancers are solely due to mutations?
Background

8. Cancer cell metabolism is linked to signaling and NADPH for rapid cell growth
Sreekumar et al 2009 Nature 457: 910
Background
Thompson & Thompson 2004 J. Clin. Onc. 22: 4217
Dang et al 2009 Nature 462: 739
Many tumors produce NADPH via glutamine
gln glu  akg  succ  fum  mal  pyr  lactate
Mutation in IDH1 in brain tumors leads to pro-oncogenic factor 2-hydroxyglutarate
a-ketoglutarate + NADPH  2HO-glutarate + NADP+
NADP+
NADPH

9. Clinical validation of cancer biomarkers
Sreekumar et al 2009 Nature 457: 910
….this was not claimed by Sreekumar et al.
Background
….this was not claimed by Sreekumar et al.
Lessons learned:
authors should disclose all data and metadata, not just graphs
biomarkers will be more robust as panel, not as single variable
validation should follow guidelines as given in the EDRN network of NCI
Debate on: urine sediment vs supernatant, normalization to creatinine vs alanine vs….)

10. Methods How many platforms do we need?
UPLC-UV-MS/MS secondary metabolites oxylipids, anthocyanins, flavonoids, pigments acylcarnitines, folates, glucuronidated & glycosylated aglycones
Twister-GC-TOF volatiles terpenes, alkanes, FFA, benzenes
nanoESI-MS/MS polar & neutral lipids UPLC-MS/MS phosphatidylcholines, -serines, -ethanolamines, -inositols, ceramides, sphingomyelins, plasmalogens, triglycerides
GCxGC-TOF primary small metabolites sugars, HO-acids, FFA, amino acids, sterols, phosphates, aromatics
350 ID
200 ID
100 ID 70
pyGC-MS monomers lignin, hemicellulose complex lipids
How many platforms do we need?
UC Davis Genome Center – Metabolomics Facility 3,000 sq.ft. 6 GC-MS, 6 LC-MS (TOFs, QTOF, FTMS, QQQ, ion traps) ~15 staff key card secured entrances, password-protected data
Methods

11. (1) Primary metabolites < 550 Da by ALEX-CIS-GC-TOF MS
20 mg breast tissue homogenization
70 eV
50-330°C ramp
-20°C cold extraction (iPrOH, ACN, water)
20 spectra/s
Methods
Dry down, derivatize to increase volatility
$60 direct costs/sample
Fiehnlab BinBase DB
Statistics
Mapping

12. (2) Volatiles < 450 Da by Twister TDU GC-TOF MS
Exhale breath on Twister
70 eV
50-330°C ramp
-70°C
20 spectra/s
400
500
600
700
Time (s)
Methods
$60 direct costs/sample HO O OH
Intensity (total ion chromatogram) O O
Fiehnlab vocBinBase DB
Statistics
Mapping

13. (1+2) Databases are critical for success
Methods

14. (1+2) Databases are critical for success
FiehnLib: Mass spectral and retention index libraries Anal. Chem. 2009, 81: 10038
1. discard poor quality signals (low signal to noise ratio )
2. cross reference multiple chromatograms
3. compound identification (mass spectra + RI matching by FiehnLib)
4. store and compare all metabolites against all 24,368 samples in 373 studies
Methods
Chemical translation service cts.fiehnlab.ucdavis.edu

15. (3) Polymers by pyrolysis GC-MS
Methods
$20 direct costs/sample
AMDIS / SpectConnect
Statistics
Mapping

16. (4) Secondary metabolites < 1,500 Da by UPLC-MS/MS
$60 direct costs/sample
Methods
target vendor software
Statistics
Mapping

17. (5) Complex lipids < 1,500 Da by nanoESI-MS/MS
$60 direct costs/sample
nanoESI infusion
chip robot
LTQ-FT-ICR-MS
High resolution
Statistics
Genedata Refiner MS
Mapping
Fiehnlab LipidBLAST
exp. MS/MS
in silico MS/MS
Methods

18. Breast Cancer: Therapeutic success depends on hormonal receptor status
lifetime risk of breast cancer in the U.S. ~ 12%
lifetime risk of dying from breast cancer 3%
in U.S., around 200k invasive plus 60k in-situ breast cancers.
in U.S., around 40k deaths by breast cancer annually.
cancer grades (1, 2, 3) reflect lack of cellular differentiation ; indicate progression
grade1
grade2
grade3
Background
In combination with surgery, endocrine therapy can treat ER+ (estrogen), PR+ (progesteron) or HER+ (Herceptin) tumors
Tumors without expression of hormone receptors (‘Triple negative’) are more likely progress to invasive states; patients have higher 5y mortality

19. Study Design
(1) Can we identify metabolites or metabolic pathways that are associated with breast cancer clinical parameters?
(2) Once we have identified those metabolic aberrations, can we validate these in a fully independent study?
First cohort 284 samples Nov normal samples
210 tumors (20 grade 1, 101 grade 2, 71 grade 3)
Second cohort
113 Samples Jan normal samples tumors (10 grade 1, 46 grade 2, 30 grade 3)
EU FP7, PI Prof Carsten Denkert, Berlin
Methods

20. Hormone receptor status vs grade
% of patients
Methods
triple neg.
Estrogen positive
Estrogen negative

21. Partial Least Square (multivariate stats)
(1) Can we identify metabolites or metabolic pathways that are associated with breast cancer clinical parameters?
Alex-CIS-GCTOF MS w/ BinBase: 470 detected compounds
161 known metabolites, 309 without identified structure.
Partial Least Square (multivariate stats)
grade1
grade2
grade3
Results
grade1
grade3
breast adipose
grade2