A Study Of Flucloxacillin Biotransformation Using Heteronuclear

Designed Biofluid Mixtures Allow

Feature-Wise Evaluation of Metabolic

Profiling Analytical Platforms

Toby Athersuch

Isaac Newton Institute for Mathematical Sciences

30/08/2011

[email protected]

MRC-HPA Centre for Environment and Health

• NIH definition:– “a characteristic that is objectively measured and evaluated as an indicator of normal biologic processes,

pathogenic processes, or pharmacologic responses to a therapeutic intervention “

Biomarkers Definitions Working Group (2001).

• Where do we find them?– Many sampling source (urine / plasma / tissue / breath / etc.)

– Combinations of measurements

• Considerations for sampling sources:– Urine Time-average systemic excretion

– Blood „Snapshot‟ of system

– Tissue Likely site of injury/concern/action/effect

• Biomarkers can potentially:– Report more specifically on type of exposure

– Give a better idea on the exposure that has actually occurred

– Reflect a variety of stages in the initiation and progression of disease, and efficacy of treatment

3

Biomarkers

• Finding biomarkers– EXPOSURE

• External exposure

• Internal dose

– BIOLOGICALLY EFFECTIVE DOSE

– EFFECT• Biological response / early biological effect

• Altered biological structure or function

• Disease onset

– SUSCEPTIBILITY• Underlying characteristics that facilitate/modulate exposure and response

• Adapted from: Committee on Biological Markers of the National Research Council. Biological markers in

environmental health research. Environ. Health Perspect. 74: 3-9, 1987.

Internal

Dose

Biologically

Effective

Dose

Early

Biological

Effect

Altered

Structure /

Function

Clinical

DiseaseExposure

Susceptibility

Biomarker

Marker progression

Influence of individual

susceptibility on progression

Key

Biomarkers

• Measuring many (10‟s – 1000‟s) of biomolecules at once– Largely untargeted approach

– (Hopefully) good coverage of molecular classes of interest (e.g. metabolites, gene transcripts, etc.)

– Minimise assumptions

– Useful for hypothesis generation – getting a list of potential candidates for further validation

• The „profile‟ may be considered as a whole– Combinations of molecules may respond in concert

– Potentially more sensitivity and specificity

• Approach across „omics‟ sciences have numerous common elements– Differ according to level/type of biological information being investigated and associated methodologies

• Generate a multiparametric dataset

• Apply multivariate statistics to derive putative biomarkers

5

Omics and Global Profiling

Metabonomics

“…the quantitative measurement of the dynamic multiparametric

metabolic response of living systems to pathophysiological stimuli or

genetic modification…” - MetabonomeNicholson JK, Lindon JC, Holmes E. 1999. 'Metabonomics': understanding the metabolic responses of living systems

to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data. Xenobiotica.

29:1181-1189.

Metabolomics

“...the complete set of metabolites / low-molecular-weight

intermediates, which are context dependent, varying according to the

physiology, developmental or pathological state of the cell, tissue,

organ or organism…” - MetabolomeFiehn O. 2002. Metabolomics – the link between genotypes and phenotypes. Plant Mol. Biol. 48(1-2) 155-171.

Metabolic Profiling

General term for „omics‟ metabolite measurements MRC-HPA Centre for Environment & Health

Some Definitions

• Profiles can be any description of the small molecule composition of a „biological

sample‟

– For each sample measure as many different molecules as possible

– Measure in as unbiased a manner as possible

– Usually derived using spectroscopy

• Can simultaneously generate (multiple) hypotheses and test them

– No a priori assumptions needed about what is important (untargeted analysis)

– Can be used to focus on specific classes or set of metabolites (targeted)

– Multivariate analysis makes this process efficient for biomarker discovery

– Combinations of biomarkers can be more sensitive and specific than one

• Profiles must be:

– Amenable to quantitative interpretation

– Reproducible (relative to „biological‟ variation)

• Profiles need not be:

– Comprehensive (limited by instrumental considerations)

– Fully resolved and annotatedMRC-HPA Centre for Environment & Health

Metabolic Profiling

PreparationCollection Analysis Data Analysis

Outline of Analysis Strategy

1. Generate profiles that reflect/report on metabolic state of biological system under study

2. Use multivariate analysis / pattern recognition to indicate profile features related to intervention/disease…

3. Identify metabolites responsible for these profile features – putative biomarkers

4. Generate testable hypotheses to validate putative biomarkers

5. Use profiles or features to derive classification models

MRC-HPA Centre for Environment & Health

Mass Spectrometry

NMR Spectroscopy Chromatography

Single quad MS

Triple quad MS

Time-of-flight MS

Ion mobility MS

Ion trap MS

FTMS

Gas chromatography

Capillary electrophoresis

UPLC

HPLC

SPE

1D NMR

Multidimensional NMR

Magic angle spinning NMR

Flow injection NMR

Capillary NMR

Hyphenation

High ThroughputPreparation Robotics

Automation

LIMS


Analytical Platforms for Metabolic Profiling

• Spectra of biofluids (urine, plasma, bile, CSF) are typically very complex, with many

overlapped peaks

• Spectral assignment is complex, but necessary in drug metabolism work and

metabonomics

H ppm

600 MHz 1H NMR spectrum of rat urine

Biofluid NMR for Metabolic Profiling

• 1D 1H NMR data amenable to analysis

– Full resolution• e.g. 32 K points / spectrum

• high computational load

• Potentially affected by minor chemical shift

variation

H ppm

– Integrated regions• Arbitrary / targeted

• Ability to incorporate moderate

chemical shift variation

• Less specific output from data analysis

Region

1 2 3 4

Sampl

e 1

Sampl

e 2

Sampl

e 3

…

Spectral Data for Multivariate Analysis

• 1D 1H NMR data amenable to analysis

– Targeted analysis• Fit model spectra of well-characterised metabolites to target spectrum

• Quantification possible through reference to internal standard

Spectral Data for Multivariate Analysis

Example: „Targeted Profiling‟ spectral fitting using Chenomx NMR Suite:

Spectrum to be fitted

Metabolite 1

Metabolite 2

Sumline


Summary

Designed Mixtures in Biofluid Analysis

Background:

Traditional „spike-in‟ experiments used for the assessment of precision and accuracy of

analyte measurements, are adversely affected by variation in the background matrix

composition. Such approaches my underestimate the risk of matrix effects confounding

accurate measurement of major metabolites

Aim:

To evaluate and illustrate how mixing intact biofluids according to a know, suitable

experimental design can help:

- evaluate individual measurements

- compare performance

- enable optimisation

… of quantitative spectral analysis tools under conditions that better approximate a real

metabonomics experiment.

Summary

From: NIST/SEMATECH e-Handbook of Statistical Methods, http://www.itl.nist.gov/div898/handbook/, Updated 6/2010

Pilot Study – Human Urine

STUDY DESIGN - Simplex Lattice

A {q, m} simplex-lattice design for q components consists of points defined by the following

coordinate settings:

The proportions assumed by each component take the m+1 equally spaced values from 0 -1,

xi = 0, 1/m, 2/m, ... , 1 for i = 1, 2, ... , q

All of the possible combinations of the proportions described are sampled in the

experimental.

STRATEGY

If a metabolite is accurately measured by targeted analysis from this data set, then the

variation in the measurements will fit a linear regression to this experimental design. In this

study we have compared a ‘targeted profiling’ approach using Chenomx NMR Suite with a

‘targeted bucketing’ approach.

http://www.itl.nist.gov/div898/handbook/

Experimental - 1


Schematic showing the experimental design.

Pooled urines taken from a previously

completed human nutritional intervention were

mixed in known proportions

Human urine

(condition A)

Human urine

(condition B)

Human urine

(control)

-10000

0

10000

-60000 -50000 -40000 -30000 -20000 -10000 0 10000 20000 30000 40000 50000 60000

t[2]

t[1]

TJA_NHLI_GA_110711_DM_SIMCA_v1.M8 (PCA-X)

t[Comp. 1]/t[Comp. 2]

Colored according to Obs ID (Primary)

R2X[1] = 0.830682 R2X[2] = 0.0653338

Ellipse: Hotelling T2 (0.95)

001b

001a

002a

002b

003a003b

004b004a

005b005a

006a

006b

007b

007a

008a008b

009b

009a

010a

010b

SIMCA-P+ 11.5 - 23/08/2011 11:39:16

Results - 1


Student (1st) attempt at preparing „real‟ biofluid samples

PCA revealed relatively weak concordance with experimental design

Experimental - 2

Pilot Study - Standard MixturesThree different standard compounds were mixed in

known proportions

Glycine Citrate

Hippurate

Results - 2

Internal

Standard

Citrate

Glycine

Pilot Study - Standard MixturesSpectra are sparse

No spectral overlap between compounds

Hippurate

Hippurate

Poor shimming Adequate shimming

Results - 2

-10000

0

10000

-20000 -10000 0 10000 20000

t[2]

t[1]

TJA_STD_GA_110810_DM_SIMCA.M8 (PCA-X)



R2X[1] = 0.574136 R2X[2] = 0.303233


001c001a

001b

002b

002a002c

003c003b003a

004a004b004c

005b005c005a

006b006a

007a007b

007c

008a008c008b

009c

009b009a

010a010b010c

SIMCA-P+ 11.5 - 23/08/2011 16:20:16

-10000

0

10000

-20000 -10000 0 10000 20000

t[2]

t[1]

TJA_STD_GA_110808_DM_SIMCA.M6 (PCA-X)



R2X[1] = 0.534172 R2X[2] = 0.295828


001c001a001b

002b002a002c

003c003b

003a

004a004b004c

005b005c

005a

006b006a

006c

007a007b007c

008a008c008b

009c009b009a

010a010b010c

SIMCA-P+ 11.5 - 23/08/2011 16:18:23

Results - 2

-14000

-12000

-10000

-8000

-6000

-4000

-2000

0

2000

4000

6000

8000

10000

12000

14000

-10000 0 10000

t[2]

t[1]

TJA_STD_GA_ALL_DM_SIMCA.M7 (PCA-X)



R2X[1] = 0.525247 R2X[2] = 0.295904


0

1

001c

005b

008a

005c

002b

010a

003c

006b

009c

004a

009b

007a

010b

002a

003b

007b

002c

004b

007c

003a

001a

010c

005a

001b

006a

009a

008c

004c

008b

006c

1001c

1005b

1008a

1005c

1002b

1010a

1003c

1006b

1009c

1004a

1009b

1007a

1010b

1002a

1003b

1007b

1002c

1004b

1007c

1003a

1001a

1010c

1005a

1001b

1006a

1009a

1008c

1004c

1008b

SIMCA-P+ 11.5 - 23/08/2011 16:41:22

Pilot Study – Standard Mixtures

Comparison of all samples by PCA revealed minimal contribution from shimming artifact

Explained by this being the result of a similar change to all measured signals.

Poor shimming

Adequate shimming

Experimental - 3

Study - Different Urines

Schematic showing the experimental design.

Three different urines with substantially different

metabolic profiles were mixed in known proportions

Rat urine

(day)

Rat urine

(night)

Human urine

Experimental - 3

EXPERIMENTAL METHODS

NMR Acquisition

Quantitative 1D 1H NMR spectra were obtained

600 MHz

NOESYPR1D pulse sequence

Sum of 128 free

Spectral width of 20 ppm.

Analytical triplicates were used for each biofluid sample

NMR Data Processing

Line broadening of 1 Hz prior to Fourier transformation.

Phase and baseline distortions

Internally to the TSP-d4 resonance relative to DSS (1H 0.00).

Targeted Profiling

Quantification of individual urinary metabolites

Chenomx NMR Suite 4.6 (Chenomx Inc., Edmonton, Canada).

Reference spectra from the Chenomx 600 MHz library

Relative concentrations of each metabolite present determined.

Targeted Bucketing

Spectral regions were defined for each metabolite of interest

The integral area of these regions was calculated

Pattern Recognition

Partial least squares regression

- Metabolite concentration data (X)

- Experimental design matrix (Y)

Goodness-of-fit (R2) and goodness-of-prediction (Q2) estimates

Targeted Fitting

Targeted Integration

Mixture

Proportions

Design Mixtures

Spectra

PLS

Results - 3

PLS

Com

pone

nt

2

PLS Component 1

12

4

5

3

7

6

98

10

SampleStudy – Different Urine

PLS scores for “Targeted Profiling”

method.

Show a relatively tight fit to the

experimental design

Samples clustered in replicates

Several metabolites that were

affected by overlap or could not be

accurately quantified identified

R2 < 0.6 = 5 of 54

R2 < 0.8 = 14 of 54


1,3

-Dim

eth

ylu

rate

1,6

-Anhydro

-ß-D

-glu

cose

N,N

-Dim

eth

ylfo

rma

mid

e

Trim

eth

yla

min

e N

-oxid

e-0.2

-0.1

-0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.01

-Me

thyln

ico

tinam

ide

2-H

yd

roxyis

obuty

rate

2-O

xo

glu

tara

te2

-Oxo

isocapro

ate

3-H

yd

roxyis

ovale

rate

3-I

ndoxyls

ulfate

Ace

tate

Ala

nin

eA

llan

toin

Arg

inin

e

Be

tain

eC

ho

line

cis

-Aco

nitate

Citra

teC

rea

tinin

eD

ime

thyla

min

e

Eth

anol

Eth

anola

min

eF

orm

ate

Fu

co

se

Fu

ma

rate

Glu

cose

Gly

cin

eG

lyco

late

Gu

anid

oaceta

te

Hip

pura

teIs

ole

ucin

eL

acta

te

Le

ucin

eM

alo

nate

Me

thanol

Me

thylm

alo

nate

Me

thyls

uccin

ate

N,N

-Dim

eth

ylg

lycin

eN

-Ace

tylg

lycin

eO

-Ph

osphocholin

e

Oxa

laceta

teP

henyla

cety

lgly

cin

eP

yru

va

te

Su

ccin

ate

Ta

uri

ne

Th

reonin

e

tra

ns-A

conitate

Trig

on

elli

ne

Try

pto

phan

Tyro

sin

eU

racil

Ure

aV

alin

eX

ylo

se

PLS model goodness-of-fit (R2 ) and goodness-of-prediction (Q2 ) metrics for individual metabolite concentrations.

Results - 3

Results - 3

Study – Different Urines

Gross urine concentration varies considerable between samples as a consequence of highly

variable dilution factor.

To remove/reduce this effect from the spectra being models, probabilistic quotient

normalisation (PQN, Dieterle et al. 2006) was applied.

The normalised spectra were regressed against the experimental design.

Remaining R2 values should therefore better reflect how well the metabolite is measured

across all samples, without the influence of overall urine dilution.

Results - 3

-0.2

-0.1

-0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1,3

-Dim

eth

ylu

rate

1-M

eth

yln

icotinam

ide

2-H

ydro

xyis

ob

uty

rate

2-O

xoglu

tara

te

2-O

xois

ocapro

ate

3-H

ydro

xyis

ova

lera

te

3-I

ndoxyls

ulfate

Aceta

te

Ala

nin

e

Alla

nto

in

Arg

inin

e

Beta

ine

Cholin

e

cis

-Aconitate

Citra

te

Cre

atinin

e

Dim

eth

yla

min

e

Eth

anol

Eth

anola

min

e

Form

ate

Fucose

Fum

ara

te

Glu

cose

Gly

cin

e

Gly

co

late

Guanid

oaceta

te

Hip

pura

te

Isole

ucin

e

Lacta

te

Leucin

e

Malo

nate

Meth

anol

Meth

ylm

alo

nate

Meth

yls

uccin

ate

N,N

-Dim

eth

ylform

am

ide

N,N

-Dim

eth

ylg

lycin

e

N-A

cety

lgly

cin

e

O-P

hosphocholin

e

Oxala

ceta

te

Phenyla

cety

lgly

cin

e

Pyru

vate

Succin

ate

Taurine

Thre

onin

e

trans-A

conitate

Trigonelli

ne

Trim

eth

yla

min

e N

-oxid

e

Try

pto

phan

Tyro

sin

e

Ura

cil

Ure

a

Valin

e

Xylo

se

1,6

-Anhydro

-ß-D

-glu

cose


Targeted Profiling

Results - 3

-0.1

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1,3

-Dim

eth

ylu

rate

1,6

-Anhydro

-ß-D

-glu

cose

1-M

eth

yln

icotinam

ide

2-H

ydro

xyis

ob

uty

rate

2-O

xoglu

tara

te

2-O

xois

ocapro

ate

3-H

ydro

xyis

ova

lera

te

3-I

ndoxyls

ulfate

Aceta

te

Ala

nin

e

Alla

nto

in

Arg

inin

e

Beta

ine

Cholin

e

cis

-Aconitate

Citra

te

Cre

atinin

e

Dim

eth

yla

min

e

Eth

anol

Eth

anola

min

e

Form

ate

Fucose

Fum

ara

te

Glu

cose

Gly

cin

e

Gly

cola

te

Guanid

oaceta

te

Hip

pura

te

Isole

ucin

e

Lacta

te

Leucin

e

Malo

nate

Meth

anol

Meth

ylm

alo

nate

Meth

yls

uccin

ate

N,N

-Dim

eth

ylform

am

ide

N,N

-Dim

eth

ylg

lycin

e

N-A

cety

lgly

cin

e

O-P

hosphocholin

e

Oxala

ceta

te

Phenyla

cety

lgly

cin

e

Pyru

vate

Succin

ate

Taurine

Thre

onin

e

trans-A

conitate

Trigonelli

ne

Trim

eth

yla

min

e N

-oxid

e

Try

pto

phan

Tyro

sin

e

Ura

cil

Ure

a

Valin

e

Xylo

se


Targeted Integration

Results - 3


Consequences of Normalisation

For the analysis of the „Targeted Profiling‟ approach, variation in most metabolites still

strongly associated with the differences between the original samples,

A greater number of metabolites with lower performance were identified:

R2 < 0.6 = 16 of 54

R2 < 0.8 = 31 of 54

Comparison of Approaches

An equivalent analysis using a „Targeted Integration‟ approach to represent the same

metabolites indicated that such a lower distribution of

R2 < 0.6 = 19 of 54

R2 < 0.8 = 38 of 54


Discussion


By being able to predict the relationship between measurements on different samples, each

with a different background, we were able to estimate the reliability of quantification –

based on explained variance (R2) - on a metabolite-by-metabolite basis, and to refine the

set of metabolites that we could safely interpret from the urinary data.

There was broad agreement between „Targeted Profiling‟ and „Targeted Integration‟

approaches.

Targeted Profiling has an advantage in quantifying metabolites with peaks in overlapped or

baseline-dominated spectral regions.

Targeted Integration has the advantage of being very rapid, and requiring minimal

repetitive, time-consuming spectral fits to be made (a process that has not been entirely

automated yet).

Using designed biofluid mixtures provides a rational basis for exploiting information from

several samples whereas spectral deconvolution is typically applied to one spectrum at a

time. Thus, the strategy may have general benefits for quantitative analysis in biofluid

metabolic profiling.


References

References

1. Weljie AM, Newton J, Mercier P, Carlson E, Slupsky C M: Targeted profiling: quantitative

analysis of 1H NMR metabolomics data. Anal. Chem. 2006. 78, 4430-4442.

2. Eriksson L, Johansson E, Kettaneh-Wold N, Wold S. Multi- and megavariate data analysis.

Principles and applications. Umetrics AB, Umea, Sweden. 2001.

3. Dieterle F, Ross A, Schlotterbeck G, Senn H: Probabilistic quotient normalization as robust

method to account for dilution of complex biological mixtures. Application in 1H NMR

metabonomics. Anal. Chem. 2006. 78, 4281-4290.

Evaluate Simple Mixtures

No matrix

Spiked into matrix

Variety of Matrices

Implementation in Future Studies

Start routinely adding in mixtures of biofluids incorporating known experimental groups

(e.g. control vs treatment A vs treatment B)

Investigate Higher-Dimensionality Mixtures

Investigate which experimental designs fit best within a routine profiling context

Suggestions welcome!


Future Work

• Metabolic profiles are information-rich phenotypes that can be characterised by

spectroscopic platforms such as NMR

• Such platforms provide information on an array of metabolites simultaneously.

• Multivariate analysis of the spectral sets can help derive biomarkers in a variety of

contexts (e.g. toxicology, efficacy, disease progression, diet, etc).

• Measurements are often made on complex biofluids with varying background

interference from the underlying matrix. Characterising how this can influence the

measurements across a sample set can help inform the analyst of their reliability.

• Using a simplex lattice design, we evaluated the performance of NMR spectroscopy

to characterise >50 common metabolites using two alternative methods for metabolite

quantification.

• We conclude that inclusion of designed mixtures in routine metabolic profiling work

may improve our understanding of which putative markers are reliable.


Summary

Biomolecular Medicine• Prof Jeremy Nicholson

• Prof Elaine Holmes

• Prof John Lindon

• Dr Hector Keun

• Dr Tim Ebbels

• Dr Muireann Coen

• Dr Jake Bundy

• Dr James Ellis

• Dr Rachel Cavill

• Dr Olaf Beckonert

• Dr Jake Pearce

• Dr Anas Kamleh

Epidemiology & Biostatistics• Prof Paul Elliott

• Prof Paolo Vineis

• Dr Anna Hansell

• Dr Claus Svendsen

• Dr David Spurgeon

• Dr Marta Blangiardo

• Dr John Gulliver

IARC• Prof Chris Wild

• Dr Mazda Jenab

• Prof Pierre Hainaut

U. Maastricht• Dr Theo de Kok

U. Utrecht• Dr Roel Vermeulen

NHRF• Soterios Kyrtopoulos

SeRMN• Dr Teodor Parella

• Dr Miriam Perez-Trujillo

• Mr Sergi Gil

Acknowledgements and Collaborators

Chenomx• Aalim Weljie

• Jack Newton

• Shahid Malik

http://www.environment-health.ac.uk

Funding and Support

Documents

A Study Of Flucloxacillin Biotransformation Using Heteronuclear