Bioinformatics for Targeted Metabolomics: Met and Unmet Needs

Klaus M. Weinberger

Biocrates Life Sciences AG, Innsbruck, Austria

3rd Annual Forum for SMEs

Information Workshop on European Bioinformatics Resources

Vienna, September 3 – 4, 2009

Agenda

• Why (targeted) metabolomics?

• Proof-of-concept in routine clinical diagnostics

• Technology platform

• Workflow integration & data analysis

• Issues

• Acknowledgements

Socrates470-399 BC

Hippocrates 460-377 BC

IntelligenceWisdom

MedicineHealth

BIOCRATES

“Creating Knowledge for Health”

... the systematic identification and quantitation of all/ biologically relevant small molecules* in a given compartment, cell, tissue or body fluid.

It represents the functional end-point of physiological and pathophysiological processes depicting both genetic predisposition and environmental influences like nutrition, exercise or medication.

* no biopolymers (nucleic acids, polypeptides)

Metabolomics is...

Why (targeted) metabolomics?

Six systems biologists examining an elephant

Transcription

Translation PTM

DNA

2.5·104

RNA

~105

Polypeptides

~106

Proteins

~107

~104

Metabolites

Enzymaticactivity

Transportetc.

Why metabolomics?

• Functional end-point of physiology and pathophysiology

• Reasonable scale of the analytical challenge

• Direct mirror of environmental influences

• (Mal-)nutrition• Exercize• Medication

Sample cohorts

Metabolic profiling(e.g. full scan LC-MS)

Differential patterninformation

Metabolomics approaches

HPLC-ToF-MS of urine samples +T O F M S : 4.995 to 9 .994 m in f rom P R01 -40-1_040092_56_1_ 02 04029486.w iff ,s aturation c orrection appl ied a=3. 56735167855777570e-004, t0= 3.08326670642854880e+ 001, subtrac ted (12.99 4 to 13.994 m in)

M ax. 32.0 c ounts.

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500m/ z, am u

0

2

4

6

8

10

12

14

16

18

20

22

24

26

28

30

32

Inte

nsity

, cou

nts

376.1168

114.0871

105. 0298

600.230 9

388.2420

327.1889

432.2672

377.1319207.1518

359.128 9134.0554

391.273018 0.0597

424.2011 584 .2329344.2083497.1240

163.1292 195.0611 570.220541 5. 23493 18.1 748297.1361

520.3189366.1678107.0682 446.2369334.1 510 624.2623

Sample: mouse urineID 0204029486

(3/8)HPLC: Waters Atlantis dC18 injection volume: 10 µl detection: pos. ToF-MS

m/z 100-1500mass accuracy: ~ 2 ppmdata content: c. 2500 features per spectrum for statistical assessment

PCA of LC/MS profiling data

Candidate drug vs. Untreated Untreated vs. Rosiglitazone

Sample cohorts

Metabolic profiling(e.g. full scan LC-MS)

Differential patterninformation

Identification of relevant metabolites

Targeted metabolomics(ID / quantitation

by SID on MS/MS)

Metabolite concentration shifts

Functional annotation

Metabolomics approaches

Pathway mapping of quantitative Mx data

Cit

ArgOrn

Argsucc

Fum

Urea

Asp

Carb-P

NO

NOS

ASL

ASS

ARG

OCT

Basic research- Functional genomics in biochemistry, physiology, cell biology,

microbiology, ecology, … Agricultural & nutrition industry

- Plant intermediary metabolism- Health effects of functional food products

Biotechnology- Optimization and monitoring of fermentation processes

Pharmaceutical R&D- Pathobiochemistry / characterization of disease models- Safety / toxicology- Efficacy / pharmacodynamics and mode-of-action

Clinical diagnostics & theranostics- Early diagnosis and accurate staging- Specific monitoring of therapeutic effects

Areas of application

History and proof-of-concept in clinical diagnostics

Sir Archibald Edward Garrod

• 1857, London – 1936, Cambridge• Educated in Marlborough, Oxford,

and London• Postgraduate studies at the AKH in

Vienna in 1884/85• Publications on chemical pathology

(e.g. of alkaptonuria, cystinuria, pentosuria)

• One gene – one enzyme hypothesis• Concept of inborn errors of

metabolism (Croonian lectures to the Royal College of Physicians, 1908)

Proof-of-concept in neonatology

• Newborn screening for inborn metabolic disorders

• replaced expensive monoparametric assays

• simultaneous detection of 40 - 60 metabolites (amino acids, acylcarnitines)

• simultaneous diagnosis of 20 - 30 monogenic diseases (AA metabolism, FATMO) with immediate treatment options

• total incidence > 1:2000

• unprecedented sensitivity, specificity, ppv

• co-pioneered in the mid-90s by BIOCRATES founder Bert Roscher

• > 1,300,000 newborns screened in Munich

• similar labs worldwide

Lessons from newborn screening

1) Quantitative tandem mass spectrometry (stable isotope dilution) is able to meet the most stringent quality criteria (precision, accuracy) for routine diagnostics

2) The concept of multiparametric biomarkers improving assay sensitivity and, particularly, specificity is valid for many monogenic (and multifactorial) diseases

3) MS-based diagnostics can save costs despite a wider analytical panel and improved diagnostic quality

Also true for therapeutic drug monitoring of immunosuppressants, antidepressants, antiretrovirals...

Goals in clinical diagnostics

Conventionaldiagnostics

genetic predisposition

healthy

latent

illMultiparametric diagnostics

• Early diagnosis• Prophylaxis instead of therapy

• Subtyping / Staging• Therapeutic drug monitoring• Phenotypic pharmacogenomics• Individualized (and more cost-

efficient) medicine

Technology, workflow integration & data analysis

• Automated extraction and derivatization

• SPE

Sample preparation

• Technical validation• Statistical analysis• Data visualization• Biochemical

interpretation

BioInformatics

1

disease2

Clinical & experimentalsamples

Diagnoses & lab data

BioBank

LIMS/Database

• Separation (LC, GC)• Quantitation (MRM, SID)• QA/QC

Analytics

Integrated technology platform

Workflow overview

Staging of diabetic and non-diabetic nephropathy by PCA-DA

MarkerViewTM

Identifying marker candidates: stage 3 vs. stage 5 kidney disease (loadings)

stage 3 stage 4 stage 5

Met

-SO

/Met

0,000

0,005

0,010

0,015

0,020

0,025

0,030

Increasing oxidative stress in progressing CKD

• Oxidation of methionine is highly indicative for oxidative stress

• Ratio of Met-SO to Met quantitative measure for this biomarker

01020304050607080901000

10

20

30

40

50

60

f(x) = 0.499538363143896 x − 5.89571444082883R² = 0.752276711067173

Metabolite vs. eGFR, non-diabetic, w/o Stage 5

ADMA (U) Linear (ADMA (U))

eGFR

Met

abo

lite

Decreasing ADMA secretion in progressing CKD

• Regression analysis to identify correlation of marker candidates with continous (clinical) variables instead of discrete (=artificial) stages

Membrane phospholipids (GPC, GPE, GPS, ...)

Lysophospholipids Free fatty acidsPUFAs

AA 20:4w6LA 18:2w6 DHA 22:6w3EPA 20:5w3

9-HODE 12-HETE 15-HETE PGD2LTB4 TXB213-HODE

SPL2

PGE2

LOX COXROS

Orchestration of fatty acid oxidation

Pathway visualization in KEGG (reference pathway)

Pathway visualization in KEGG (human)

Dynamic pathway visualization in MarkerIDQ

Exploring ‚metabolic shells‘ around metabolites

Route finding between metabolites across pathways

Reactions vs. Reactant pairs!

Issues I: Databases

Parallel / competing initiatives with incompatible / proprietary data formats KEGG MetaCyc, HumanCyc, etc. Reactome HMDB OMIM Lipidomics consortia ...

Compartmentalization not well depicted Incompleteness / generic entries (phospholipids, acylcarnitines,

etc.) Lack of curation Lack of publication

Standardization Instrument vendors oppose common data formats What meta-data to record? No valid guidelines for quantitation of endogenous metabolites

(FDA guidance was developed for xenobiotics) Nomenclature vs. analytical reality (sum signals, isomers, etc.)

Normalization Absolute quantitation overcomes the need for analytical

normalization Role of sample types (plasma, CSF, urine, tissue homogenates,

cell extracts, ...) How can biological normalization work? Are there ‚house-

keeping metabolites‘?

Issues II: Standardization and normalization

Overfitting & correction Suitable clustering algorithms for multivariate data sets? Metabolites are no equivalent independent variables

Analytical validity/variability are usually not considered Often, groups of metabolites are synthesized or degraded

by the same enzyme(s) Consecutive reactions within a pathway/network depend on

each other (flux analysis!) How to incorporate this in biostatistics? Weighting? Derived

parameters, ratios, etc.? How to exploit this in (automated) plausibility checks?

Issues III: Biostatistics

Summary I

• Metabolomics depicts the functional end-point of genetics and environment

• Targeted metabolomics data are analytically reproducible and allow immediate biochemical interpretation

• Proof-of-concept has been achieved in routine diagnostics of inborn errors of metabolism

• Many metabolic biomarkers are valid across species and enable translational research

• Comprehensive targeted metabolomics bridges the gap to open profiling approaches

Summary II : Success factors for biomarker development

Validated quantitative

assays

Well-documented biobanking

Patent strategy and experience

Clinical & scientific experts

Biochemical plausibility &

understanding

Solid multi-variate

biostatistics

Biomarker candidates

Diligentstudy design

Validated biomarkers

Selectedpartners

Acknowledgements

BioinformaticsDaniel Andres Olivier LefèvrePaolo Zaccaria Florian BichtelerMarc Breit Manuel GoglBernd Haas Mattias BairRobert Eller Hamza Ovacin

Gerd Lorünser Yi Zao

AnalyticsStefanie Gstrein Sascha DammeierHai Pham Tuan Cornelia RöhringTherese Koal Ali AlchalabiVerena Forcher Ines UnterwurzacherStefan Urban Doreen KirchbergRalf BogumilPatrizia HoferLisa KörnerPeter Enoh

Statistics & BiochemistryIngrid Osprian Marion BeierVera Neubauer Oliver LutzMatthias Keller Denise SonntagHans-Peter Deigner Ulrika Lundin

Admin, IT & BizDevBrad Morie Anton Grones Ingrid SandnerDoris Gigele Georg Debus Wolfgang SamsingerElgar Schnegg Patricia Aschacher