Multivariate Profiling of Metabolites in Human Disease $(function(){PrimeFaces.cw("Tooltip","widget_formSmash_items_resultList_29_j_idt799_0_j_idt801",{id:"formSmash:items:resultList:29:j_idt799:0:j_idt801",widgetVar:"widget_formSmash_items_resultList_29_j_idt799_0_j_idt801",showEffect:"fade",hideEffect:"fade",target:"formSmash:items:resultList:29:j_idt799:0:fullText"});});

Multivariate Profiling of Metabolites in Human Disease $(function(){PrimeFaces.cw("Tooltip","widget_formSmash_items_resultList_29_j_idt799_0_j_idt801",{id:"formSmash:items:resultList:29:j_idt799:0:j_idt801",widgetVar:"widget_formSmash_items_resultList_29_j_idt799_0_j_idt801",showEffect:"fade",hideEffect:"fade",target:"formSmash:items:resultList:29:j_idt799:0:fullText"});});
Multivariate Profiling of Metabolites
in Human Disease
Method evaluation and application to prostate cancer
Elin Thysell
Department of Chemistry
Umeå University, Sweden 2012
Copyright © Elin Thysell
ISBN: 978-91-7459-344-0
Front cover picture by Petrus Sjövik Johansson
Elektronisk version tillgänglig på http://umu.diva-portal.org/
Tryck/Printed by: VMC-KBC Umeå
Umeå, Sweden 2012
To my family
i
ii
Abstract
There is an ever increasing need of new technologies for identification of molecular markers for
early diagnosis of fatal diseases to allow efficient treatment. In addition, there is great value in
finding patterns of metabolites, proteins or genes altered in relation to specific disease
conditions to gain a deeper understanding of the underlying mechanisms of disease
development. If successful, scientific achievements in this field could apart from early diagnosis
lead to development of new drugs, treatments or preventions for many serious diseases.
Metabolites are low molecular weight compounds involved in the chemical reactions taking
place in the cells of living organisms to uphold life, i.e. metabolism. The research field of
metabolomics investigates the relationship between metabolite alterations and biochemical
mechanisms, e.g. disease processes. To understand these associations hundreds of metabolites
present in a sample are quantified using sensitive bioanalytical techniques. In this way a unique
chemical fingerprint is obtained for each sample, providing an instant picture of the current
state of the studied system. This fingerprint or picture can then be utilized for the discovery of
biomarkers or biomarker patterns of biological and clinical relevance.
In this thesis the focus is set on evaluation and application of strategies for studying metabolic
alterations in human tissues associated with disease. A chemometric methodology for
processing and modeling of gas chromatography-mass spectrometry (GC-MS) based
metabolomics data, is designed for developing predictive systems for generation of
representative data, validation and result verification, diagnosis and screening of large sample
sets.
The developed strategies were specifically applied for identification of metabolite markers and
metabolic pathways associated with prostate cancer disease progression. The long-term goal
was to detect new sensitive diagnostic/prognostic markers, which ultimately could be used to
differentiate between indolent and aggressive tumors at diagnosis and thus aid in the
development of personalized treatments. Our main finding so far is the detection of high levels
of cholesterol in prostate cancer bone metastases. This in combination with previously
presented results suggests cholesterol as a potentially interesting therapeutic target for
advanced prostate cancer. Furthermore we detected metabolic alterations in plasma associated
with metastasis development. These results were further explored in prospective samples
attempting to verify some of the identified metabolites as potential prognostic markers.
Keywords
Metabolite profiling, metabolomics, predictive metabolomics, mass spectrometry, GC-MS,
biomarkers, chemometrics, design of experiments, multivariate data analysis, prostate cancer,
bone metastases, plasma
iii
iv
Sammanfattning på svenska
Grunden för liv utgörs av kemiska processer som med ett samlingsnamn benämns metabolism
eller ämnesomsättning. Tack vare dessa processer kan celler växa, föröka sig, underhålla sin
struktur och anpassa sig till förändringar. De små molekylerna i kroppen som innefattas av
metabolismen kallas metaboliter. Aminosyror, fettsyror, socker och hormoner är exempel på
klasser av sådana metaboliter. Det forskningsområde som fokuserar på systematiska studier av
metaboliter och förändringar i desamma kallas för metabolomik.
Metabolomik eller global metabolitprofilering avser att detektera metabola fingeravtryck i
biologiska prover (t.ex. blodplasma, urin eller vävnad) genom analys med känsliga bioanalytiska
instrument. Bakgrunden till metodiken är att halter av metaboliter förändras enligt specifika
mönster vid exempelvis sjukdom eller miljöpåverkan. Dessa mönster kan liknas vid ett unikt
fingeravtryck karaktäristiskt för det studerade systemets fysiologiska tillstånd. Trenden inom
metabolomikområdet går mot användandet av allt mer känsliga analytiska tekniker, vilka
skapar stora mängder informationsrika data. Till de mest använda analytiska teknikerna hör
kärnmagnetisk resonansspektroskopi (NMR), gaskromatografi kopplat till masspektrometri
(GC-MS) och vätskekromatografi kopplat till massspektrometri (LC-MS). De komplexa
fingeravtrycken som dessa tekniker genererar, bestående av halterna av hundratals metaboliter,
analyseras sedan med kemometriska eller multivariata analysmetoder (t.ex. PCA, PLS och
OPLS) för att skapa tolkningsbara kartor över förändringar i metabolismen.
Syftet med studierna i denna avhandling var att utveckla metoder för att kunna använda
metabolomik inom medicinska frågeställningar med målet att öka möjligheterna att
diagnostisera, förstå och i förlängningen behandla allvarliga sjukdomar.
För att skapa förutsättningar för detektion och identifiering av metaboliter har vi inom vår
forskningsgrupp utvecklat en strategi
för kurvupplösning kallad hierarkisk multivariat
kurvupplösning (H-MCR). Förenklat är kurvupplösning en matematiskt metod som gör det
möjligt att kvantifiera och identifiera metaboliter som inte kan separeras analytiskt genom
exempelvis kromatografi. Vi har i detta arbete visat att i kombination med multivariata
analysmetoder gör H-MCR det möjligt att skapa system för tolkning av metabola processer,
detektion och identifiering av biomarkörer eller biomarkörmönster samt prediktion av
oberoende prover. Dessutom visar resultaten att detta kan göras tidseffektivt i stora provserier.
De utvecklade metoderna har använts i studier som syftar till att förstå de mekanismer som
ligger bakom att vissa fall av prostatacancer utvecklas och blir aggressiva och livshotande för
patienten, medan andra förblir långsamväxande och ofarliga. Ofta är det närvaron av
dottertumörer, metastaser, som bestämmer allvaret vid en cancersjukdom. Inom ramen för
denna avhandling försöker vi förstå vad som reglerar tillväxt av benmetastaser vid aggressiv
prostatacancer för att i förlängningen förstå hur dessa bör förebyggas och behandlas. Det
övergripande syftet är att utveckla känsliga metoder för att hitta tumörer och för att förutsäga
v
hur farliga de är, så att patienter kan få rätt behandling i ett botbart skede av sjukdomen.
Förhoppningsvis kan dessa markörer även bidra till utvecklingen av nya behandlingsmetoder
för prostatacancer.
För att behandla patienter med avancerad prostatacancer är det mycket betydelsefullt att hitta
metastaser tidigt i utvecklingen. Metastaser från prostatatumörer återfinns ofta i skelettetet. Vi
har därför analyserat normal prostatavävnad och vävnad från benmetastaser från patienter med
prostatacancer och andra cancersjukdomar. På detta sätt hittades entydiga mönster som
utmärker benmetastaser hos patienter med prostatacancer. Något överraskande upptäcktes
höga halter av kolesterol i metastasvävnad från patienter med prostatacancer, i jämförelse med
de andra undersökta vävnaderna och i jämförelse med metastasvävnad från andra cancertyper.
Vidare visade studien att tumörcellerna både verkar kunna bilda kolesterol på egen hand och ta
upp det från omgivningen. Baserat på våra resultat i kombination med tidigare fynd tror vi att
kolesterol används till olika processer som är viktiga för en tumörcell, till exempel för att den
ska kunna växa och invadera omkringliggande vävnad. I studien hittades även höga nivåer av
andra metaboliter som förekommer vid ämnesomsättning i kroppens celler och vävnader. Bland
dessa återfanns bland annat metaboliter som möjligen även går att spåra i blod hos patienter
med metastaser.
Idag ställs diagnosen prostatacancer först och främst genom att man mäter halten av PSA, ett
prostataspecifikt antigen i blod. Ett problem är dock att PSA inte kan användas för att skilja på
aggressiv cancer och benign sjukdom. I dag kan ett positivt PSA-prov betyda många saker, allt
ifrån en infektion, inflammation eller förstorad prostata till i värsta fall en aggressiv cancer.
Målet är att hitta något som är lika enkelt som PSA-provet, dvs detekterbart i ett blodprov, men
som kan tala om vilka patienter som bär på den aggressiva formen av sjukdomen och är i behov
av behandling. I dag överbehandlas många patienter samtidigt som andra förbises. Genom att
studera metabola fingeravtryck i vävnads- och blodprover från friska och patienter kan vi lära
oss mer om bakomliggande mekanismer och förhoppningsvis utveckla och ge mer individuellt
anpassade behandlingar på sikt.
vi
List of papers
This thesis is based on the following papers, which are referred to in the text by their
Roman numerals. In the list of papers Thysell E and Johansson ES refers to the same
person.
I.
Jonsson P, Johansson ES, Wuolikainen A, Lindberg J, Schuppe-Koistinen I,
Kusano M, Sjostrom M, Trygg J, Moritz T, Antti H, Predictive metabolite
profiling applying hierarchical multivariate curve resolution to GC-MS data - A
potential tool for multi-parametric diagnosis. Journal of Proteome Research
2006, 5(6):1407-1414.
II.
Thysell E, Pohjanen E, Lindberg J, Schuppe-Koistinen I, Sjöström M, Moritz T,
Jonsson P, Antti H, Reliable profile detection in comparative metabolomics.
OMICS, A Journal of Integrative Biology 2007, 11(2): 209-224.
III.
Thysell E, Chorell E, Svensson M B, Moritz T, Jonsson P, Antti H, Processing of
mass spectrometry based metabolomics data for large scale screening studies
and diagnostics. Submitted manuscript 2011.
IV.
Thysell E, Surowiec I, Hörnberg E, Crnalic S, Widmark A, Johansson A I,
Stattin P, Bergh A, Moritz T, Antti H, Wikström P, Metabolomic Characterization
of Human Prostate Cancer Bone Metastases Reveals Increased Levels of
Cholesterol, PLoS ONE, 2010, 5(12): e14175
V.
Thysell E, Stattin P, Moritz T, Wikström P, Antti H, Evaluation of metabolic
alterations in patient plasma associated with disease aggressiveness in prostate
cancer, Manuscript 2011.
Paper I and II are reprinted with permission from the American Chemical Society and Mary
Ann Liebert, Inc. respectively
vii
Other papers by the author not appended
in the thesis
VI.
Pohjanen E, Thysell E, Jonsson P, Eklund C, Silfver A, Carlsson I B, Lundgren
K, Moritz T, Svensson MB, Antti H, Multivariate Screening Strategy for
Investigating Metabolic Effects of Strenuous Physical Exercise in Human Serum.
Journal of Proteome Research 2007, 6(6): 2113 -2120.
VII.
Andersson C D, Thysell E, Lindström A, Bylesjö M, Raubacher F, Linusson A, A
Multivariate Approach to Investigate Docking Parameters' Effects on Docking
Performance. Journal of Chemical Information and Modeling, 2007, 47(4):
1673 -1687.
VIII.
Pohjanen E, Thysell E, Lindberg J, Schuppe-Koistinen I, Moritz T, Jonsson P,
Antti H, Statistical multivariate metabolite profiling for aiding biomarker pattern
detection and mechanistic interpretations in GC/MS based metabolomics.
Metabolomics, 2006, 2(4): 257-268.
IX.
Skytt Å, Thysell E, Stattin P, Stenman U-H, Antti H, Wikström P, SELDI-TOF
MS versus prostate specific antigen analysis of prospective plasma samples in a
nested case-control study of prostate cancer. International Journal of Cancer,
2007, 121(3): 615-620.
X.
Lindahl C, Simonsson M, Bergh A, Thysell E, Antti H, Sund M, Wikström P,
Increased Levels of Macrophage-secreted Cathepsin S during Prostate Cancer
Progrogression in TRAMP Mice and Patients. Cancer Genomics Proteomics
2009, May-Jun, 6(3): 149-159.
viii
Notations
The following notations will be used in this thesis. Matrices are represented by bold,
capital letters e.g. C and vectors are denoted by bold, lower case letters, e.g. p.
Vectors are column vectors unless stated otherwise.
C
Matrix of weight vectors for Y, [MxA]
E
Residual matrix of predictor variables, [NxK]
F
Residual matrix of response variables, [NxM]
P
Matrix of loading vectors for Y, [KxA]
T
Matrix of score vectors for X, [NxA]
X
Matrix of predictor variables, [NxK]
Y
Matrix of response variables, [NxM]
p
Loading vector for X, [Kx1]
t
Score vector for X, [Nx1]
p
predictive
o
orthogonal
ix
Abbreviations
ATP
CoA
CRPC
CV
CYP11,
CYP17
DA
DIMS
DNA
DoE
FTIR
GC
HSD3B2
H-MCR
IGF1
LC
MS
MVA
NIPALS
NMR
mRNA
OSC
OPLS
PCA
PLS
PSA
RT-PCR
SVD
TIC
TOF
adenosine triphosphate
coenzyme A
castration-resistant prostate cancer
cross validation
cytochrome P450, family 11
cytochrome P450, family 17
discriminant analysis
direct infusion mass spectrometry
deoxyribonucleic acid
design of experiments
fourier transform infrared spectroscopy
gas chromatography
hydroxy-delta-5-steroiddehydrogenase , 3 beta- and steroid delta-isomerase 2
hierarchical multivariate curve resolution
insulin-like growth factor 1
liquid chromatography
mass spectrometry
multivariate data analysis
non-linear iterative partial least squares
nuclear magnetic resonance spectroscopy
messenger ribonucleic acid
orthogonal signal correction
orthogonal projections to latent structures
principal component analysis
partial least squares
prostate specific antigen
reverse transcription polymerase chain reaction
single value decomposition
total ion current
time of flight
x
Table of Contents
Abstract
iii
Sammanfattning på svenska
v
List of papers
vii
Notations
ix
Abbreviations
x
Table of Contents
xi
Background
2
Metabolomics
3
Metabolomics in prostate cancer research
6
Multivariate data processing and analysis
8
Hierarchical multivariate curve resolution
9
Chemometrics
11
Predictive metabolomics
14
Aims of the study
17
Results
18
Paper I
18
Paper II
20
Paper III
22
Paper IV
24
Paper V
26
Conclusion and future perspectives
28
Acknowledgements
31
References
33
xi
1
Background
What can you learn from a simple blood test during a routine medical checkup? For instance you can be told if you have an infection, determine your
mineral content, and monitor or assess the effectiveness of a drug. The level
of glucose in serum can be used to diagnose diabetes, elevated levels of
human chorionic gonadotrophin confirm early pregnancy, and the level of
creatinine in serum provides information about renal function. Imagine, if
you could also diagnose cancer, at an early stage of development, with a
blood test within minutes.
The physiological and biochemical composition of a blood sample can be
regarded as a signature of biomarkers, reflecting the biological state of an
individual. When the balance in a living system is disturbed it will be
reflected in the biological components of that system. Biomarkers are
naturally occurring components of the cell that can be used as indicators of a
specific physiological state, such as disease or other aspects of health.1 The
clinical applications of biomarkers are, for example, disease detection,
identification of people at risk for developing disease, detection of recurrence
after therapy, or to guide personalized treatment.2-4
2
Cancer is a term used for diseases where transformed cells divide without
control and are able to invade other tissues (The National Cancer Institute,
NCI5). Over recent years, knowledge of the complex processes that act to
overrule normal biological regulation during cancer progression has
increased dramatically. Today, one of the great challenges in cancer research
is to gain mechanistic understanding of cell transformation that can further
lead to detection of suitable biomarkers for cancer. Such biomarkers could
represent e.g. altered patterns of gene expression, inflammation,
hyperplasia, or hyperproliferation.6 Detecting cancer at as early stage as
possible is directly related to the effectiveness and outcome of the treatment
of the patient.7 However, many of the existing markers are not sensitive
enough to detect early stages, but instead work well for the detection of late
stage tumors, and prognostic information is often definitive only for patients
with an already fatal disease.8-10 In order to work also for the detection of
cancer at early stages, biomarkers need to be released in to the circulation in
measurable amounts early in disease progression, even by a small
asymptomatic tumor. It is also important that the level of the marker
molecule is not affected by non-cancer disease, since the ability to detect
cancer is then compromised greatly. It is also crucial that the marker have a
high specificity for the tissue of origin. If not, it is likely that the levels in
healthy individuals produced by other tissues will be high and overlapping
with the levels measured in patients.11 The discovery of biomarkers for early
disease detection in cancer is extremely difficult and researchers are putting
a lot of effort and resources into the discovery of cellular components that
can be used to diagnose disease.12-15
This thesis focuses on development and application of strategies for studying
metabolic alterations in human tissues and plasma associated with disease.
The developed strategies are applied to the study of prostate cancer, with the
aims to identify possible prognostic metabolite markers associated with
aggressive disease and development of bone metastases. The resulting
metabolites or metabolite patterns could be of value as novel biomarkers but
also as providers of new mechanistic insights for prostate cancer
progression.
Metabolomics
At the same time as cells grow and divide, they consume and produce small
chemical molecules known as metabolites. Metabolites are often referred to
as low molecular weight compounds (<1 kDa) involved in chemical reactions
that occur inside the cells of living organisms to uphold life, i.e. the
metabolism. The chemical diversity of the metabolome, defined as the
3
complement of all detectable metabolites, is large and includes a wide range
of compound classes, e.g. carbohydrates, amino acids, organic acids, sterols,
nucleosides (Figure 1). The quantity and number of metabolites vary with
changing conditions such as environment, diet and in response to
disease.16,17 The research field of metabolomics seeks to understand the
relationship between expressed metabolites in human tissues and biological
mechanisms by studying differences between samples in relation to their
metabolic composition. In order to detect these differences, the hundreds to
thousands of metabolites that are present in a sample need to be quantified.
This is done using sensitive bioanalytical techniques. In this way a unique
chemical fingerprint is generated for each sample that represents the
metabolic composition of the sample.
Figure1 Metabolites representative for the chemical diversity of the metabolome.
Metabolites are end products of the hierarchy starting with genes (genome)
and ranging to the collection of gene transcripts (transcriptome) and
proteins (proteome) (Figure 2). In the field of systems biology, information
from ’omics’ analyses is combined to elucidate the interactions between
genes, proteins and metabolites.18,19 Although genomics, transcriptomics,
proteomics, and metabolomics should be considered as complementary
techniques, metabolomic profiles are regarded as containing integrated
information about the events that take place at different levels in the
organism as a result of their particular location downstream of the genome,
transcriptome and proteome.20-22 The collection of metabolites in a sample,
i.e. the sample metabolome, is highly dynamic, changing over time. Global
metabolic profiling by metabolomics, therefore, provides a direct picture of
the current state and phenotype of an organism, allowing discovery of
clinical biomarkers or biomarker patterns. The size of the metabolome varies
from a few hundred to a few thousand metabolites depending on the
4
organism studied (not including lipids).23,24 In addition, over 100 000
molecules are believed to be present in humans due to consumption of food,
drugs etc.25 In recent times, metabolomics has evolved from a conventional
profiling technique into one used to study biological systems as an
interacting system of genes, proteins, metabolites and cellular events. By
combining information from genes, proteins and metabolites, global models
of biosystem function can be generated.26 This global analysis can help to
further identify possible diagnostic and prognostic biomarkers as well as
uncover altered pathways and thereby understand disease processes.27-29
Metabolomics analyses involve several important steps before a reliable
biological interpretation can be carried out. These steps include planning the
experiment, sampling, sample handling and preparation, chemical analytical
analysis, data processing, statistical analysis or modeling, and validation.
The output of this chain of events is imperative for drawing correct
conclusions about the biological system under investigation.30 Hence, many
of these steps need to be carefully standardized, optimized, and/or validated.
This includes standardized protocols for sampling, sample handling and
instrumental analysis, optimized strategies for sample preparation, and
validation of statistical models. Guidelines on reporting of studies and
methods have been recently suggested by the Metabolomics Standardization
Initiative (MSI).31-33
Figure 2. Summary of hierarchical levels in the organization of the cell involved in the
progression from genotype (hereditary information) to phenotype (structure and function).
5
Metabolomics in prostate cancer research
Cancer is one of the leading causes of death in economically developed
countries and the incidence is increasing as a result of population aging and
an adoption of lifestyle choices associated with cancer, such as smoking, diet
and physical inactivity.34 Prostate cancer is one of the most common cancers
in the Western world and roughly 10 000 men are diagnosed with the
disease every year in Sweden according to Swedish official statistical data
(Socialstyrelsen35). For those diagnosed with prostate cancer, prediction of
clinical outcome is primarily performed by digital rectal examination,
transrectal ultrasound guided biopsy and measurement of serum levels of
prostate specific antigen (PSA). If increased PSA levels are detected, prostate
biopsies are taken and cancer diagnosis is confirmed or refuted
microscopically. At present, most of the men diagnosed have PSA levels
between 4-10 ng/mL. In this range, the specificity of PSA testing is only 2030 %.36 The low specificity is due to the fact that PSA is prostate, not prostate
cancer, specific and leaks into the circulation not only due to cancer but also
during inflammation, prostatitis and benign prostate hyperplasia (BPH). As
a result many men that undergo a prostate biopsy examination will not have
prostate cancer. To reduce unnecessary biopsies, tests that can rule out
cancer alone or that can increase specificity in combination with PSA are
needed.37-40
The prognosis of prostate cancer in early stages is highly variable. A large
group of men may have tumors that are highly unlikely to cause symptoms
for many years. If such indolent tumors could be identified at the time of
diagnosis, many men could be spared unnecessary treatments. On the other
hand, about one fourth of the cases suffer from aggressive, fatal cancer
(Socialstyrelsen) eventually spreading to distant sites in the body. Current
diagnostic methods cannot separate indolent tumors from the aggressive
forms, nor predict metastatic prostate cancer at a curable stage of the
disease. The standard therapy for patients with advanced prostate cancer is
to reduce the amount of circulating androgens, either by surgical or medical
castration. Castration therapy decreases androgen receptor signaling and
causes decreased tumor cell proliferation and increased tumor cell apoptosis
41, which is probably preceded by decreased blood supply to the tumor 42.
After a period of initial remission tumors relapse and by that time they are
no longer responsive to castration treatment, and are so termed castrationresistant prostate cancer (CRPC). Therapeutic options for patients with
CRPC are limited and in general the best outcome for these patients is to
maintain, or to improve, their quality of life.
6
Limitations of current diagnosis methods have led to an intense focus on the
potential of molecular biomarkers to improve both diagnostics and
individual prognostics.
When cancer cells grow their metabolism is altered and optimally the
metabolites that are produced could be detected and used as biomarkers for
the disease. Tumors cells are known to have different metabolic profiles
compared to normal cells. These general properties include, for instance,
increased aerobic glycolysis, production of lactate and higher bioenergetic
demands.43-46 Metabolic alterations occur early in the neoplastic
transformation of prostate cells. Over the past decade, changes in choline
phospholipid metabolism have been associated to aggressive forms of cancer
including brain, prostate and breast cancers (reviewed by Glunde 47). Choline
serves as precursor for cell membrane phospholipids and as a methyl donor
in DNA methylation and DNA repair mechanisms48, however the complete
role of choline metabolism in prostate tumorigenesis is still unknown.
Prostate cells have a unique metabolite profile among human organs by
producing high concentrations of the major components in prostate fluid;
citrate, PSA, polyamines e.g. spermine and myo-inositol.49,50 In the process
of abnormal and uncontrolled growth, prostate cells shift from accumulating
and secreting citrate into an active oxidation state where chemical energy is
generated from citrate through the citric acid cycle by the production of ATP.
51
The increased level of ATP is used to nourish the large bioenergetic
demand in malignant prostate cells. Another metabolic change related to
high proliferation rate found in malignant cells is increased lipid
biosynthesis, used to produce building blocks for membrane formation and
intercellular signaling. This is achieved, for example, by increased synthesis
of acetyl CoA from citrate. Acetyl CoA can then be used as a precursor in
lipogenesis and cholesterolgenesis.52 Key enzymes involved in the synthesis
of cholesterol and fatty acids have been reported to increase activity in
prostate cancer cells.53,54
The major interest in the potential application of metabolomics for
identification of biomarkers for prostate cancer followed a recent paper by
Sreekumar and colleagues55. In this study the metabolite composition in
tissue, urine, and plasma from prostate cancer patients was profiled using
liquid (LC) and gas chromatography (GC) – time of flight mass spectrometry
(TOFMS). They identified the amino acid sarcosine as a potential marker for
prostate cancer cell invasion, migration, and aggressiveness. A number of
recent studies have collectively indicated that mass spectrometry-based
methods could be used to characterize metabolomic changes during cancer
progression and to further identify possible diagnostic and prognostic
7
biomarkers, or biomarker patterns, as well as increase our knowledge about
disease progression.56-62 Metabolic profiling of blood plasma, or serum, has
been used to unveil metabolic alterations associated with prostate
cancer.55,62-65
Multivariate data processing and analysis
In metabolomics, techniques that separate molecules by way of their varying
size and charge are often used to characterize the metabolites present in a
sample. The resulting chemical fingerprint should ultimately represent the
complete sample metabolome. However, metabolites come from many
different compound classes and are present in a large range of
concentrations (Figure 1). Thus, there is not one single analytical technique
that can cover the whole metabolome. Instead many complementary
analytical tools must be used in order to get the most comprehensive
representation possible.
Today the most frequently used methods for analysis of the metabolome are
nuclear magnetic resonance (NMR) spectroscopy and hyphenated
techniques such as gas chromatography (GC) and liquid chromatography
(LC) coupled to mass spectrometry (MS). In addition, Fourier transform
InfraRed spectroscopy (FTIR) have been used together with direct infusion
mass spectrometry (DIMS) as well as other complementary techniques
(reviewed by Dunn66). NMR and FTIR require minimal sample preparation,
however, detection limits are higher compared to the MS-based techniques
and elucidation of spectra composed of many metabolites can be
problematic. For that reason, hyphenated techniques are generally preferred
to allow both quantification and identification of as many metabolites as
possible. In the selection of a specific analytical method it has to be assessed
whether the method of choice is suitable for answering the particular
biological question. However, in metabolomics studies it is often not known
which metabolites are the most interesting beforehand, or at what
concentrations they are present. Therefore, general and more global
analytical methods are often used initially, followed by validation by means
of targeted analysis using more high resolution techniques.
GC-MS has become an important analytical platform for generating
metabolic fingerprints and comparing samples in global metabolite
analysis.58,67-70 The reason for its frequent application is due to its relatively
high sensitivity and reproducibility, but particularly because identification of
detected metabolites by mass spectral databases is rather straightforward.
However, the requirement to derivatize non-volatile metabolites increases
the sample preparation time and often complicates the identification of
8
unknown compounds.71 The GC method used in this thesis relies on
derivatization by oximation followed by silylation prior to analysis, in order
to provide coverage of the largest range of metabolites. Detected compound
classes include alcohols, aldehydes, amino acids, amines, organic acids,
sugars, sugar acids, sugar amines, sugar phosphates, purines and
pyrimidines.30 The efficiency of derivatization is an important factor as it has
an effect on the reproducibility of the analyzed metabolites. Isotopically
labeled standards, eluting with an even spread over the chromatogram, are
added to each sample to monitor extraction and derivatization effects.
Before statistical analysis the analytical data needs to be processed so that
the same identity is assigned to the same metabolite in each sample. The
generation of high quality data for further statistical analysis, interpretation,
and identification is central in the search for new markers as well as for a
deeper understanding of underlying biological processes. In this method, it
is the quality of the information that will be decisive for successful results
and a key issue is to create data that are representative for the metabolic
composition of the studied samples. For hyphenated data this can be
achieved by data deconvolution or curve resolution.72 Curve resolution can
be seen as a mathematical method for generating chromatographic profiles
and mass spectra for the detected metabolites, as well as enhancing
analytical resolution. The chromatographic profiles are then used for
quantification, while the corresponding mass spectra are used to identify the
metabolites, either by means of spectral database comparisons or by de novo
interpretation.
Hierarchical multivariate curve resolution
Hierarchical
multivariate
curve
resolution
(H-MCR)73
extracts
chromatographic profiles and mass spectra from GC-MS data of many
biological samples simultaneously, as compared to other curve resolution
methods that are performed on an individual sample basis.74,75 The
improvements using this approach are that no matching of resolved
compounds is needed and that the same compound will be quantified
identically over all samples, using the shared spectral profile. Another
advantage with the H-MCR method is that completely overlapping
compounds, eluting at identical retention times, can be resolved as long as
the concentration ratio between them and their mass spectra differs.
The H-MCR method includes the following processing steps: (1) smoothing
of each sample in all mass channels by a moving average; (2) alignment by
finding maximum covariance between the samples total ion current (TIC)
chromatograms; (3) division of data in to time windows; and (4) multivariate
9
curve resolution72 of each time window separately. This generates a
chromatographic profile for each compound in each sample with a
corresponding common spectral profile (Figure 3). Next, a data matrix X is
created that can be used for any kind of multivariate data analysis (MVA) or
other statistical analysis with the aim of comparing samples. In the data
matrix each row represents one sample and each column represents one
compound. The value in each cell of the matrix represents the integrated
area under a resolved chromatographic profile in one sample.
(a)
Raw chromatographic profiles
WINDOW 4
(c)
(b)
Set time windows
Aligned chromatographic profiles
WINDOW 4
Chromatographic profiles
Mass spectral profiles
6
1
2
3
4
5
6
3
12
4
5
WINDOW 4
FIGURE 3. An overview of the main steps included in H-MCR. (a) Smoothing of mass channels
and alignment of sample chromatograms. (b) Division into time windows, edges set at low
intensity points. (c) Each time window is then resolved separately by H-MCR, resulting in a
chromatographic and corresponding mass spectral profile for each resolved compound.
An extension to the H-MCR method also showed that this curve resolution
could be carried out predicatively for independent sample sets (Paper I),
which can prove to be of great value in biomarker pattern verification and
validation as well as various important applications, such as diagnosis and
high-throughput analysis (Papers II-V).
10
Chemometrics
Chemometrics is the computational field for planning experiments and
extracting information from high-dimensional data, i.e. the information
aspect in the study of complex systems. 76-78 Chemometrics can be seen as a
toolbox containing statistical methods that can handle many dependent
variables, so that multivariate thinking can be present in every step from
setting up studies to verifying the results. Per definition, there are two main
branches of chemometrics i) Design of experiments (DOE) and ii)
Multivariate data analysis (MVA). By combining DOE and MVA, an efficient
tool for studying systems that include multiple variables and correlated
responses is created. One important issue when dealing with human subjects
is the need to minimize confounding variation or, at best, being able to
detect and handle this variation. Here, the chemometric methodology can be
a valuable contribution together with previous experience and knowledge
related to studies.
Design of experiments (DOE)
The main aim of DOE is to maximize the information output from a
minimum number of experiments.79,80 In DOE, variation is introduced into
the data in a systematic fashion so that the effects of investigated variables
(for example sex, age and disease) and their interactions on one or many
responses (e.g. response to drug treatment) can be deciphered by means of
statistical analysis. DOE has been utilized in metabolic studies for
optimization of analytical procedures and for detecting statistically
significant changes and related biomarker patterns.81-83 Study design is
indispensable in biomarker discovery and also pointed out to be a main
reason why biomarkers fail to be accepted for clinical use.84
In this work, DOE was used (i) for diversity-based selections of
representative sample subsets (Papers I, II, III and V), (ii) to investigate
and optimize the outcome of the H-MCR method (Paper II) and (iii) to
create a dataset consisting of standard compounds that was varied in
concentration(Paper II).
Multivariate data analysis (MVA)
Multivariate data analysis is the branch of chemometrics concerned with
analysis and interpretation of complex data structures. Multivariate
projection methods for data exploration (e.g., principal component analysis,
PCA)85,86 and regression (e.g., partial least squares, PLS)87,88 have been
shown to be very useful for interpreting the systematic changes that exist
11
between many samples characterized by the relative concentrations of a
multitude of metabolites.89-94 Several features are associated with
multivariate projection-based model systems, such as separation of
systematic variation from noise, managing many and correlated variables,
outlier detection, handling of missing values, possibilities for model
validation, and prediction of independent samples. These characteristics
have been specifically attractive for pushing the global metabolite analysis
forward in terms of understanding complex interactions in biological
systems.95-98
PCA
Principal component analysis (PCA) is the most widely used, unsupervised
projection method for exploratory analysis and data compression. This
central multivariate data analysis tool extracts the systematic variation in the
original data matrix X by a reduced set of latent variables. A latent variable
is a concealed variable that cannot be directly measured or observed, but is
described by other measured variables.99 Latent variables can for instance,
be hidden metabolite structures in a blood sample related to gender, age or
disease. The two most commonly used methods for calculating a PCA model
are non-linear iterative partial least squares (NIPALS)91 or single value
decomposition (SVD)92 algorithms. The NIPLS algorithm is used throughout
this work due to its ability to handle missing values that are often present in
large metabolomics datasets.
X=TPT + E
The PCA algorithm compresses the systematic information in the original
data X in to a latent variable representation, TPT, so that it explains the
largest amount of variation in the original data. The scores (T) gives an
overview of the samples and how they relate to each other. Interpretation of
the patterns in the scores is found in the corresponding loadings (P). The
loadings express how each variable contributes to the separation among
samples and reveal relative importance of each variable. Unsystematic
information, e.g. noise, is stored in the residual matrix, E. PCA gives a
summary of the data and essential systematic effects can be easily visualized
in a separate latent variable and interpreted individually.
In this work PCA has been used to: (i) provide an overview of the data and
for detection of deviating samples (Papers I-V), (ii) create a basis for
selections of representative sample subsets in the compressed multivariate
space (Papers I and III) and (iii) predict analytical replicates, revealing
good analytical conditions (Paper III).
12
PLS and OPLS
In metabolomics the requirement to transform complex data in to
knowledge is reached by the integration of chemistry, biology and medicine
with statistics. The supervised regression method partial least squares (PLS)
is a particularly important data analysis tool for these kinds of applications.
PLS extracts latent variables by maximizing the explained variation in X as
well as the covariation between two sets of data i.e. X and Y.
The extension of PLS i.e., orthogonal PLS methodology (OPLS)100, has
greatly improved the interpretation of large complex datasets, such as those
generated within global metabolite analyses.77,101,102 OPLS is a specialized
PLS algorithm, where the systematic variation in data can be separated into
correlated (predictive) variation, TpPpT and uncorrelated (orthogonal)
variation ToPoT. For single-y OPLS models, the multidimensional variation
in the metabolite profiles is compressed into a single dimension that
maximizes the differences between groups, for instance patients and
controls. When Y is qualitative, e.g. denoting class membership, the OPLS
method is called OPLS discriminant analysis (OPLS-DA).103 For two classes,
the model representation produces a predictive score vector, tp that capture
between class variation, and the uncorrelated (orthogonal) score vectors, To
capture any within-class variations. The sources of orthogonal variation is
often related to specific study conditions such as sample handling, sample
storage, experimental problems, instrument drift, age, gender and
environmental factors.
X = TpPpT + ToPoT + E
Y = TpCpT + F
In metabolomics studies the number of variables is generally high compared
to the number of samples. To reduce the probability to find correlations by
chance, methods such as cross validation, permutation test and external test
sets for validation are applied.104-106
OPLS is the selected regression method for multiple sample comparisons
and sample predictions in the work included in this thesis. OPLS has been
used to correlate the metabolic information against the day of sample
collection in a study of rat urine toxicity and to classify aspen leaf extracts
and human blood plasma samples (Paper I), (ii) model and the predict
systematic metabolic variation related to the acute effect of strenuous
exercise (Paper III) (iii) extract and interpret the systematic variation in
tissue and plasma profiles related to prostate cancer disease progression
(Papers IV-V).
13
Predictive metabolomics
Metabolomics has attracted great interest as a sensitive technique for
elucidation of biomarkers, or biomarker patterns, in biofluids or tissues
associated with disease-related processes. However, even though many
recent results have provided proof of the ability of the technique, there are
still challenges ahead in order to move from isolated study results to
predictive or diagnostic systems based on verified marker patterns.
Among the main challenges identified for the future of metabolomics are; i)
diagnosis/prognosis, ii) result verification, iii) screening of large sample sets,
and iv) linking metabolic pattern changes to mechanistic information. To
meet these challenges a number of issues need to be addressed including
e.g., generation of high quality data in a high throughput fashion,
development of predictive systems for sample characterization and
classification and elucidation of robust biomarker patterns of biological
relevance. Key factors in this are of course the availability of representative
samples, as well as sensitive and robust analytical characterization.
However, issues such as study design, data processing and data analysis are
equally important in order to address the aforementioned challenges.
Predictive metabolomics is a concept developed in our lab, by applying
chemometrics to metabolomics studies as a means to help in this
development towards a more comprehensive utilization of the metabolomics
technique in solving biological questions. The concept involves strategies for
sample selection, planning of studies and development of analytical
protocols all based on DOE, followed by strategies for data processing, data
analysis and validation based on multivariate statistical analysis (Figure 4).
Today we have a predictive system generating high quality data in a high
throughput fashion based on GC-TOFMS data (Paper I)107. The crucial step
in this development was the extension to the H-MCR algorithm making it
possible to resolve data from independent samples predictively. Thus, by
connecting predictive H-MCR processing to multivariate projection analysis
it became possible to obtain a predictive chain of events including both data
processing and sample characterization, or classification. As a consequence,
validation of all steps in the process could be accomplished but more
importantly, diagnostic modeling based on high quality metabolite profiles
also became possible.
14
1)
DESIGN OF EXPERIMENTS
– to maximise information output
Sample collection, preparation and chemichal characterisation according to:
STANDARDIZED PROTOCOLS
HIERARCHICAL MULTIVARIATE CURVE RESOLUTION
-to resolve overlapping peaks
Samples
1
1
Metabolites
Model samples:
Resolved data
n
1
1
Samples
2)
Metabolites
n
PREDICTION:
New samples
processed by
model sample
H-MCR parameters
New samples:
Resolved data
p
m
3)
MULTIVARIATE ANALYSIS
– Sample comparison and prediction
PREDICTION:
New samples
processed by
model sample
MVA -parameters
to1
t1
4)
VALIDATION:
- Pattern verification and Biological interpretation
po1
to1
p1
tp1
Samples
Metabolites
Figure 4. Schematic diagram detailing the predictive metabolomics strategy. (1) Design of
experiments and standardized protocols should be used throughout the study (2) Model
samples (all or a representative selection of samples) are resolved by H-MCR. Obtained HMCR parameters can then be used to resolve new samples or remaining samples not
selected in the H-MCR processing. (3) The resulting data obtained from the H-MCR
processing and treatment according to the H-MCR parameters is saved in two separate data
tables. (4) Sample comparisons by means of multivariate data analysis can then be carried
out by modelling the processed data and the model can be used to make predictions of the
treated samples. Alternatively, the processed and treated data can be merged and used for
calculating a model based on all samples.
15
The additional benefit of predictive H-MCR, being much faster compared to
the original H-MCR processing, enabled the processing phase to become
high throughput. This allowed screening of large sample sets without
compromising the data quality, something that potentially will be highly
valuable for future applications, e.g. in screening large populations for
specific diseases.
Since the introduction of the predictive metabolomics concept is has been
applied to a number of studies with medical or clinical focus. These include
applications to cancer64,91, exercise and nutrition105,108,109
and
neurodegenerative disease.90,110
16
Aims of the study
The overall aims of the studies included in this thesis were to:
Evaluate a chemometric methodology for processing and modeling of GCMS based metabolomics data with the ambitions of generating a predictive
system for; i) generation of representative data, ii) validation and result
verification, iii) diagnosis and iv) screening of large sample sets.
Apply the methodology to studies of prostate cancer in humans to;
i) understand disease development and ii) reveal metabolic markers for
disease aggressiveness.
17
Results
Paper I
Predictive metabolite profiling applying hierarchical multivariate
curve resolution to GC-MS data - A potential tool for multiparametric diagnosis
This study presents an extension of the hierarchical multivariate curve
resolution (H-MCR) method that makes it possible to import new
independent samples into a predictive framework for processing of
metabolic GC-MS data and subsequent multivariate modeling. Application of
the strategy is proven to facilitate result validation and verification in terms
of biological interpretation, as well as metabolite changes in new
independent samples, e.g. analytically characterized at a different point in
time. As a consequence of this, diagnostic modelling and metabolic screening
of large sample sets based on high quality data is made feasible.
Within an organism or cell, it is often combinations of metabolites that are
responsible for the variation of interest. In these situations it is essential to
be able to use the whole metabolite profile to extract biomarker patterns, as
opposed to a single biomarker. Predictive models of metabolite pattern
changes in biological systems will potentially be of great value with regard to
18
early disease diagnosis, and disease prognosis, as well as in clinical
monitoring for personalized treatment and healthcare.
When comparing metabolite compositions in different samples, the resolved
compounds from one sample need to be identified in the other samples, if
they exist. The matching of resolved compounds in new samples that have
not been part of the curve resolution procedure by H-MCR includes the
following steps: Smoothing of the data in the same way as the model
samples; alignment of samples using the target for the model samples;
division of the data into time windows using the same edges; resolving the
new data for each time window, using the spectral profiles found in the
model samples.
We show that the proposed strategy is efficient for describing metabolic
changes in many different biological sample matrices. Here, presented for
data sets of aspen leaf extracts from a plant development study, rat urine
samples from a toxicity study, and human blood plasma samples from male
and female subjects. By this we have achieved greatly reduced times for data
processing, since only a smaller representative subset is processed by curve
resolution. In addition it addresses problems with biased classification
modeling in cases where one class contains more samples than another.
Using the proposed strategy, the same number of samples from defined
classes can be processed and modeled simultaneously, whereas remaining
samples can be predicted and thus not affect the classification in any way.
For the strategy to work optimally it is important to use experimental design
introducing all known sources of variation into the model and to optimize
analytical procedures to create reproducible data over time and between
labs. A requirement for achieving reliable prediction result is clearly that the
diagnostic information in the deciding model is founded on biologically
relevant variation for the study of interest. Additionally, it is vital that the
selection of diverse samples for building models includes variation that is
representative for future prediction samples. Methodologies for selecting a
representative set of samples intended for building models is presented and
investigated in Paper III.
19
Paper II
Reliable profile detection in comparative metabolomics
In global metabolite profiling the quality of the metabolic information is key
in finding biomarker patterns that are indicative of a specific physiological
status. Low quality data are more prone to produce spurious correlations
resulting in over-fitting and misinterpretation. This is problematic in the
search for diagnostic marker patterns that ought to be verified over multiple
studies and ultimately used for predictions in new samples. For these
purposes, representative data, reflecting the biology of the system in a
reliable fashion is a requirement in order for global metabolite analysis to
become a tool for clinical diagnosis.
As previously described, all compounds in a complex sample cannot be
separated on the GC column. Thus, signals for different compounds will be
recorded at the same time, resulting in a mix of them in the mass
spectrometer. Methods for curve resolution are often used to separate
overlapping metabolites. However, for complex samples this is not always
straightforward. Instead, profiles will be estimates of the pure compounds,
whereas others will be left unresolved. These issues can occur because of e.g.,
difficulties with minor compounds vs. noise, high concentration differences
for a single metabolite in different samples and errors in the estimation of
the number of metabolites to be resolved. To deal with these problems the
curve resolution method H-MCR is controlled by several factors, which can
be adjusted by the user.
In this paper, a strategy for optimizing and understanding the data
processing is suggested with the overall aim to produce a framework for
generating representative data with regards to both quantitative and
qualitative issues. The considered characteristics of the resolved GC-MS data
were the number of resolved compounds, the quality and reproducibility of
the resolved chromatographic profiles, and the quality of the resolved mass
spectra. These characteristics were investigated in relation to the user
adjustable factors by DOE. As a first step, all combinations of the factor
settings were studied in a 24 full factorial design and correlated to responses
describing the above characteristics. This systematic analysis provided
knowledge about the effects of adjustable factors, as well as on how to select
settings beneficial for different data characteristics.
20
Framework:
GC-MS analysis of biological samples and analytical replicates
Definition of factors affecting the outcome of the processing and
key characteristics describing different properties of the data
Design of experiments (DOE) in defined factors
H-MCR processing according to DOE protocol and quantification of
data characteristics
Multivariate analysis relating changes in factors to quantitative
and qualitative characteristics of the output data
Optimization of data processing to create representative data
By altering the H-MCR factor settings in a systematic way we show that the
possibilities for extracting data with different characteristics are extensive.
The results show that by only using the number of resolved metabolite
profiles as the response to optimize against, the quality of the output data is
partly neglected. This will induce negative effects on issues such as sample
comparisons, predictions, and metabolite identifications being the
cornerstones of metabolomics. Thus the importance of producing high
quality data cannot be underestimated.
By considering both quantitative and qualitative features of the data it was
possible to understand and optimize the performance of the H-MCR method.
In this way a framework for resolving a high number of metabolite profiles
exhibiting both high chromatographic and spectral quality was obtained.
This is a general strategy that can be applied to any type of data processing
provided that the important processing parameters and key output data
characteristics can be defined. The results, however, are not general,
meaning that the optimal settings will vary depending on the data,
suggesting that the presented approach should be used as an integrated step
in global metabolite profiling.
21
Paper III
Processing of mass spectrometry based metabolomics data for
large scale screening studies and diagnostics
Efficient screening of large sample sets, or sample banks, for informative
patterns of metabolites is an issue of major concern for the progression of
metabolomics. From our point of view, efficient metabolomic screening
should take advantage of the higher sensitivity of the analytical techniques to
create data of higher quality by using sophisticated data processing. To do
this in a high throughput fashion the strategy needs be based on the selection
and use of representative samples relevant for the question at hand. By
selecting representative sample subsets by chemometric approaches and
using these subsets to perform curve resolution (H-MCR) and multivariate
classification analysis (OPLS-DA), it was possible to form a predictive
metabolomics screening strategy for GC-TOFMS data on human blood
serum samples. The samples were collected in a study examining the effect of
strenuous physical exercise. Healthy and regularly training male subjects
performed four identical tests of strenuous ergometer cycling exercise. Blood
samples were collected before and immediately after each exercise session,
to improve the understanding of human metabolism in connection to acute
physical exercise. The selection of sample subsets was performed according
to two different principles, the first being a selection based on property data
and the other being based on already acquired analytical data processed
using a fast and crude processing method69.
t3
t1
Representa ve sample subset
Predic on sample subset
t2
Figure 5. Visualization of the representative sample subset selection based on analytical data
(marked as black spheres and remaining test samples as gray spheres). The subset was
selected by space-filling design which maximizes the minimum Euclidean distance between
111
the nearest neighbors of the selected observations .
The results showed that the presented strategy provide an organized
approach, which could be applied to efficient screening of biobanks, or other
22
large sample sets, while retaining data quality and interpretation. Also, the
methodology could be applied to verify biomarker patterns in independent
sample sets, in this case, analytically characterized eight months later than
the model samples, which is the basis for working and validated diagnostic
systems. The time and utility for producing representative data is very
important for an efficient screening of large sample sets. Generally, curve
resolution methods are limited in the number of samples that can be
processed. This is mainly due to time-consuming processing and limitations
in computer capacity. However, the predictive feature of the H-MCR method
solved this issue as processing of 16 model samples took 6h and 29 min.,
while predictive processing of remaining 77 samples took merely 10 min.
This shows that as long as the selected subset is representative in terms of
retained variation and that the predictions can be performed in a robust and
reliable way, this is an efficient strategy for producing high quality data, with
no limitations concerning the number of samples that can be treated.
The proposed method can be used for:
- Sample bank mining where sample availability is constrained and it
is very important to extract as much information as possible from a
limited number of samples.
- High throughput screening of large sample sets producing high
quality data for interpretation and biomarker identification.
- Fast and high quality metabolic screening to help select samples for
more expensive and time consuming analyses in other types of
-omics studies.
- Developing systems for biomarker pattern verification, being an
extremely important issue from a clinical perspective and for
validation of findings in independent studies.
Even though the results are promising there are still challenges that need to
be addressed to reach a complete and robust method for screening and
predicting samples over time. For example, this study includes a
homogenous human cohort, which is not representative for the whole human
population. This is clearly something that needs to be considered in, e.g.
disease diagnosis modeling. Also, it would be valuable to further evaluate the
strategy by performing a completely independent study, applying the same
test to another population and make predictions for these subjects into the
existing model. Another issue that needs to be investigated is strategies for
continuous updating of models to assure robust and reliable end results.
Also, of utmost importance is the continuous investigation and optimization
of protocols and quality control of sample handling and analytical
characterization.
23
Paper IV
Metabolomic characterization of human prostate cancer bone
metastases reveals increased levels of cholesterol
In this study we applied the predictive metabolomics strategy to characterize
metabolomic alterations associated with prostate cancer progression. The
overall aim was to identify possible prognostic metabolite markers
associated with aggressive disease and bone metastases, which could,
furthermore, increase our understanding about disease progression.
The study was performed with the hypothesis that potential markers for
aggressive prostate cancer could be found by discovering metabolites at high
levels in bone metastases and then investigate if these factors also were
increased in primary tumours and in blood samples from patients with
metastatic disease. We performed a GC-TOFMS-based metabolomics study
of prostate cancer bone metastases in comparison to corresponding normal
bone, primary prostate cancer tumours and normal prostate tissue. The
tissue samples were obtained from patients operated for metastatic spinal
cord compression or pathologic fractures at Umeå University Hospital
2003–2009.112 In addition, characterized blood samples from patients with
and without diagnosed bone metastases were included, to identify
metabolites that could be used to improve prognostication and aid in therapy
of advanced prostate cancer. The findings in the bone metastasis tissue were
verified in a separate test set, also including metastatic bone tissue from
cancers of different origin.
The result showed that prostate cancer bone metastases have a different
metabolic signature, compared to normal bone and to bone metastases from
other cancers. Among the detected metabolites we specifically noted the high
levels of cholesterol when differentiating prostate cancer bone metastases
from normal bone tissue and from other bone metastases. Cholesterol has
previously been linked to prostate cancer disease progression (reviewed by
Solomon113) and was, therefore, selected for further analysis.
In order to examine possible causes for the high levels of cholesterol in
prostate cancer bone metastases, paraffin embedded tissue sections were
immunostained for enzymes involved in the influx of exogenous cholesterol
to the cell, as well as de novo synthesis. The results showed that prostate
cancer bone metastases have the possibility for both uptake and the de novo
synthesis of cholesterol. Other studies have also indicated that increased
availability of cholesterol may be of relevance for the growth and
24
development of bone metastases, including apoptosis, cell proliferation,
migration and invasion.114-117
In a resulting study, cholesterol synthesis was targeted in an in vitro model
for bone metastases118. The aim was to further understand the role of
cholesterol in the interaction between prostate cancer cells and bone and it
was shown that cholesterol is important for prostate cancer cell growth in
vitro and in co-culture with bone. In the process of tumor growth, prostate
cancer cells induced a lytic response in bone and release of the insulin-like
growth factor 1 (IGF1), which acts as a powerful survival factor.119
Interestingly, it was further shown that by targeting cholesterol synthesis
and the IGF1R simultaneously, prostate cancer cell growth was more
efficiently inhibited than by either therapy alone. As a result, inhibition of
IGFR1 in combination with existing apoptosis-inducing treatments, such as
statins, castration and chemotherapy are now being studied in animal
models for their effects on prostate cancer bone metastases. Taken together,
the presented results and previous findings in the literature indicate the
prospect of using cholesterol inhibitors as treatment or chemopreventive
agents for prostate cancer metastases. 113 However, novel drugs are then
probably needed that efficiently target peripheral organs.
We also hypothesized that cholesterol, by its conversion in to androgens via
metabolic enzymes, possibly increases androgen receptor signaling as well as
castration resistant tumor growth in patients treated with androgendeprivation therapy.120,121 This did not however, seem to be the case in a
subsequent study examining the pathways involved in synthesis of
androgens from cholesterol using gene expression arrays, RT-PCR, and
immunohistochemistry. In this study, neither of the steroid-converting
enzymes in the early steps of cholesterol conversion into testosterone;
CYP11, CYP17, or HSD3B2, showed significantly higher levels in CRPC bone
metastases compared to non-castrated (hormone-naïve) bone metastases
(Jernberg E et al., unpublished data).
In addition we discovered metabolic differences between primary prostate
tumor tissues from high-risk patients, with and without established bone
metastases. Finally, differentiating metabolite profiles in blood plasma from
patients diagnosed with high risk tumors, with and without established bone
metastases could be identified. The pattern in blood plasma associated to
aggressive disease is further examined in Paper V.
25
Paper V
Evaluation of metabolic alterations in patient plasma associated
with disease aggressiveness in prostate cancer
At the time when the prostate cancer has spread outside the prostate organ
and metastasized, only palliative therapies are available. In this study,
predictive metabolomics was used to identify metabolites in clinical series of
plasma samples associated with high risk prostate cancer and biochemical
relapse after radical prostatectomy. The findings could possibly aid in the
search for new biomarkers for detection of clinically significant tumors at a
curable time-point.
Multivariate data analysis were carried out with the strategy of first
discovering plasma metabolites associated with prostate cancer disease
progression in a series of patients stratified according to prostate cancer risk
groups (The National Comprehensive Cancer Network. Practice Guidelines
in Oncology-Version.1.2010. Prostate cancer), including patients at five
different stages ranging in disease severity from benign to metastatic
disease, and secondly, by investigating these metabolites for association with
biochemical relapse after radical prostatectomy in a second series of
patients. The fact that detected metabolic markers were evaluated in two
different sample cohorts facilitated cross study verification for relevance and
robustness.
13 metabolites were highlighted as important in OPLS-DA models including
all detected plasma metabolites both for i) difference between patients with
benign and metastatic disease and ii) difference between sample groups
related to prostate cancer risk. Within the extracted metabolite patterns for
metastatic disease two metabolites were detected which were possibly
consumed by aggressive prostate cancer (decreased plasma levels with
increased PCa risk and increased after surgery in the relapse but not in the
non relapse group). We further found increased levels of four metabolites in
plasma from patients with metastatic disease, while no supposed tumorderived metabolite was detected at an early stage of the disease (increased
plasma levels in low to intermediate prostate cancer risk and decreased
levels after surgery). In addition, verification of metabolite markers for
metastatic disease detected previously by us and others was performed. By
this we were able to confirm decreased plasma levels of stearic acid and
increased levels of pseudouridine with metastatic disease, while others were
not confirmed or not detected (e.g. phenylalanine, taurine, glutamate) or
detected (e.g. sarcosine).
26
An interesting finding was also that we could identify differences related to
biochemical relapse by evaluation of multivariate metabolite patterns before
and after radical prostatectomy. The results highlighted an interesting
phenomenon that by introducing an intervention to the biological system, in
this case the surgery, metabolic differences between the relapse and nonrelapse groups seemed to be magnified. This is a methodologically
interesting event that might be of assistance in diagnostic or prognostic
modeling and in gaining mechanistic understanding of metabolic events
such as variations in disease sub groups.
In this study the majority of the metabolites pointed out as interesting
markers in the different comparisons were unknowns. This highlights the
fact that identification of metabolites is still an issue in metabolomics
studies. Without comprehensive spectral libraries the usefulness of the data
is limited. However, strategies for identification of unknown metabolites will
be applied in terms of fractionation, parallel spectroscopic analyses and
structural determination by NMR.
27
Conclusion and future perspectives
This thesis describes research directed towards solving current challenges in
metabolomics including result verification, screening of large sample sets
and working applications for disease diagnosis and prognosis.
The presented chemometric methodology for processing and modeling of
GC-MS based metabolomics data creates a predictive system for sample
characterization and classification as for linking changes in metabolic
patterns to biological mechanisms. In order to meet the aforesaid challenges
and to obtain essential conclusions about the biological system under
investigation, methods for study design, data processing and data analysis
are essential. Furthermore, standardized protocols to ensure sensitive and
robust analytical characterization and availability of representative samples
are key components.
The strategy can be used for efficient screening of biobanks, or other large
sample sets, while retaining data quality and interpretation. Also, the
methodology could be applied to verify biomarker patterns in independent
sample sets, analytically characterized later in time than the model samples,
which is the basis for working and validated diagnostic systems. Hence, the
strategy can be used to create predictive models for biomarker pattern
verification in biological systems that potentially will be of great value for
early disease diagnosis and prognosis, as well as in clinical monitoring for
personalized treatment and healthcare.
28
Metabolomics has several advantages when studying biological systems.
Primarily, the metabolome describes the actual functional status related to
the phenotype of the organism. Secondly, metabolomics in view of global
metabolite profiling is a hypothesis free method, and therefore unexpected
relationships and metabolite reactions can be discovered, which in itself is
hypothesis generating.
One of the main issues for improvement of metabolomics is how to provide a
complete coverage of the metabolism of a cell, tissue, organ or organism in
order to get a mechanistic understanding of the biological systems. Today,
no bioanalytical technique can generate a complete metabolomic description
of even the simplest organism. However, only by combining technologies will
we ever have the means to increase our coverage of the metabolome. Thus,
there is a great need for improved methods for data integration in order to
cross correlate the information produced by a variety of analytical
approaches. Here chemometrics will have a great impact in developing
strategies that efficiently can handle the large amounts of data of great
complexity.
The challenge of metabolite identification has been known for several years,
and has drawn attention to the need for comprehensive metabolite
databases. In this way changes in identified metabolite levels can be related
to global changes across classical metabolic pathways, so that significant
understanding is reached.
Global metabolite profiling in its current form can efficiently be used as an
exploratory tool to find interesting areas of the metabolism that can be
further investigated by targeted analysis. This was presented in paper IV
where we applied the predictive metabolomics strategy to characterize
metabolomic alterations associated with prostate cancer progression. We
discovered elevated cholesterol as one interesting metabolic marker for the
development of bone metastases in prostate cancer. The finding has made us
examine effects of cholesterol and cholesterol depletion on growth of
prostate cancer and its metastases, and will hopefully lead to improved
treatment of bone metastases in the future. Our study was the first to report
on metabolomic characterization of prostate bone metastases in humans.
Furthermore, we applied the predictive metabolomics strategy to detect new
biomarkers for detection of clinically significant tumors by finding
metabolites in plasma associated with high risk prostate cancer and
biochemical relapse after radical prostatectomy. The metabolite markers
suggested by us to be associated with aggressive prostate cancer in paper IV
were to some extent verified in paper V (stearic acid and pseudouridine),
while others were not. Furthermore, additional interesting metabolites were
29
detected in paper V. These metabolites need to be identified as well as
carefully verified in separate patient cohorts to allow further interpretation
of the results.
In conclusion, this work has confirmed the potential of metabolomics for
finding new diagnostic and prognostic markers and more importantly to
increase mechanistic knowledge about disease progression in general and for
prostate cancer in particular.
Gaining mechanistic understanding about cancer and finding sufficient
diagnostic tools is a problem that needs to be addressed in a multilevel
fashion. Hopefully, the work included in this thesis has brought us one step
closer to the utopia of diagnosing cancer with a simple blood test.
Now this is not the end.
It is not even the beginning of the end.
But it is, perhaps, the end of the beginning.
-Sir Winston Churchill
30
Acknowledgements
Jag vill tacka alla som på ett eller annat sätt bidragit till att min avhandling
nu är klar.
Ett stort och innerligt tack till...
Henrik Antti, min huvudhandledare, som inspirerat, gett mig trygghet och alltid
kommit med nya ideer. Tack för att du genom kontrollerad frihet gett mig möjlighet
att utvecklas och för att du alltid visat mig ett stort förtroende. Tack också för din
positivt smittsamma inställning till livet i allmännhet och till vår forskning i
synnerhet. Äntligen är det dags att sparka ut din första doktorand.
Johan Trygg och Michael Sjöström, mina biträdande handledare. Det har varit
skönt att veta att jag har haft er i bakfickan. Ni har stor del i kemometrins framgång
och jag är tacksam för den kunskap ni givit mig.
Pernilla Wikström, min sist i raden tillkomna biträdande handledare. Du är en
sann förebild och jag vill speciellt tacka dig för att du berikat mitt arbete med
spännande forskning och ovärdelig kunskap.
Thomas Moritz, för all din hjälp och för att du tagit dig tiden att svara på alla mina
frågor. Tack till övriga hjälpsamma medarbetare på UPSC; Krister Lundgren, IngaBritt Carlsson, Annika Johansson och Izabella Surowiec som alla bidragit till
analyserna av proverna i det presenterade studierna.
Min kära medarbetere och vänner: Anna Wuolikainen, Elin Chorell, Lina
Mörén, Elin Näsström, Pär Jonsson, Tommy Öhman, Carl Wibom, Emma Jernberg,
Annika Nordstrand, Mattias Hedenström, Max Bylesjö, Hans Stenlund, Anton
Lindström, Fredrik Petterson, David Andersson, Anna Linusson för ett mycket gott
sammarbete och trevliga möten. Stort tack även till er som korrekturläst
avhandlingen. Ett särskilt tack till Jonathan Gilthorpe som sett över engelskan. Ett
jättestort tack till Anna och Lina, för att ni bidragit med diverse energitillskott när
jag behövt det som mest i form av mat, fika och skratt! Lill-Elin, det känns tryggt att
veta att gruppen kommer fortsätta sin tradition av att samla på Elinar. Tommy, för
härliga pratstunder och för din omtänksamhet.
Pär, för att du alltid tar dig tid att lyssna och för din goda förmåga att lösa problem.
Jag har lärt mig så mycket av dig. Du skulle vara förmögen om du fått en krona för
varje gång jag vänt mig om och sagt, -Duuu Pär.....? Tack för alla svar!
31
Nuvarande och forna kollegor, som bidragit till att göra kemiska institutionen
till en trevlig arbetsplats och inspirerande forskningsmiljö. Barbro Forsgren och
Carina Sandberg och Lars-Göran Adolfsson för att ni alltid gett mig hjälpsamma svar
på allt ifrån administrativa frågor till datorer. Speciellt tack till Patrik Andersson och
Mattias Hedenström för mycket trevliga uppföljningsmöten, som skett i sista sekund
vart eviga år.
Mia och Elin, vart ska jag börja. Stöpta i samma form, kokade i samma gryta,
grillade i samma ugn och tillsammans utgör vi en perfekt komponerad måltid. Ni är
oskattbara!
Mina kära vänner, Louise, Simon, Erik, Ylva, Tobbe, Johan, Theresia, Linda,
Andreas, Cathis, Hans, Stina, Simon, Anna, Fredrik med familjer. Tack för många
trevliga stunder och för att ni visar att vi inte är för gamla för galna upptåg.
Min älskade släkt, som gjort att jag varje söndag haft ett kalas att se fram emot.
Det är underbart att ha er så nära och ni är ovärderliga för mig.
Våra härliga grannar, Carlsons och Wennerholms som återkommande visat
omtanke och bidragit med ingredienser till våra barns pannkakor.
Ralf, för att du är en klippa och alltid ställer upp med allt ifrån kulinariska middagar
till diverse layouter och trycksaker.
Matilda, min älskade systeryster, för otaliga barnvakter med kort varsel och mysiga
stunder.
Bror Petrus, för att du påminner mig om att allt går att ordna och att ingenting är
omöjligt (i sann morfar-anda).
Sofie, för att du är ett härligt tillskott i familjen och för att din siluett pryder mitt
omslag.
Mamma och pappa, för att ni är min trygga hamn och för att ni alltid ställer upp.
Mina käraste, Agaton, Vendela och Marcus, jag älskar er!
32
References
1.
Atkinson, A.J., et al. Biomarkers and surrogate endpoints: Preferred
definitions and conceptual framework. Clin. Pharmacol. Ther. 69,
89-95 (2001).
2.
Canney, P.A., Moore, M., Wilkinson, P.M. & James, R.D. Ovariancancer antigen CA125 - A Prospective clinical-assessment of its role
as a tumor-marker. British Journal of Cancer 50, 765-769 (1984).
3.
Anderson, J.L., et al. Randomized trial of genotype-guided versus
standard warfarin dosing in patients initiating oral anticoagulation.
Circulation 116, 2563-2570 (2007).
4.
Kiss, I., et al. Colorectal cancer risk in relation to genetic
polymorphism of cytochrome P450 1A1, 2E1, and glutathione-Stransferase M1 enzymes. Anticancer Res. 20, 519-522 (2000).
5.
(NCI), T.N.C.I. http://www.cancer.gov/. (2011).
6.
Srinivas, P.R., Kramer, B.S. & Srivastava, S. Trends in biomarker
research for cancer detection. Lancet Oncology 2, 698-704 (2001).
7.
Narain, V., Cher, M.L. & Wood, D.P. Prostate cancer diagnosis,
staging and survival. Cancer and Metastasis Reviews 21, 17-27
(2002).
8.
Duffy, M.J. Serum tumor markers in breast cancer: Are they of
clinical value? Clin. Chem. 52, 345-351 (2006).
9.
Tomlins, S.A., et al. Urine TMPRSS2:ERG Fusion Transcript
Stratifies Prostate Cancer Risk in Men with Elevated Serum PSA.
Science Translational Medicine 3(2011).
10.
Meinhold-Heerlein, I., et al. An integrated clinical-genomics
approach identifies a candidate multi-analyte blood test for serous
ovarian carcinoma. Clinical Cancer Research 13, 458-466 (2007).
11.
Diamandis, E.P. Cancer Biomarkers: Can We Turn Recent Failures
into Success? J. Natl. Cancer Inst. 102, 1462-1467 (2010).
12.
Srivastava, S. & Kramer, B.S. Early detection cancer research
network. Lab Invest 80, 1147-1148 (2000).
33
13.
Pepe, M.S., et al. Phases of biomarker development for early
detection of cancer. J. Natl. Cancer Inst. 93, 1054-1061 (2001).
14.
Srinivas, P.R., Srivastava, S., Hanash, S. & Wright, G.L. Proteomics
in early detection of cancer. Clin. Chem. 47, 1901-1911 (2001).
15.
Etzioni, R., et al. The case for early detection. Nature Reviews
Cancer 3, 243-252 (2003).
16.
Holmes, E., et al. Human metabolic phenotype diversity and its
association with diet and blood pressure. Nature 453, 396-U350
(2008).
17.
Yap, I.K.S., et al. Metabolome-Wide Association Study Identifies
Multiple Biomarkers that Discriminate North and South Chinese
Populations at Differing Risks of Cardiovascular Disease
INTERMAP Study. Journal of Proteome Research 9, 6647-6654
(2010).
18.
Kitano, H. Systems biology: A brief overview. Science 295, 16621664 (2002).
19.
Kussmann, M., Raymond, F. & Affolter, M. OMICS-driven
biomarker discovery in nutrition and health. Journal of
Biotechnology 124, 758-787 (2006).
20.
Fiehn, O. Metabolomics - the link between genotypes and
phenotypes. Plant Molecular Biology 48, 155-171 (2002).
21.
Kell, D.B. Metabolomics and systems biology: making sense of the
soup. Current Opinion in Microbiology 7, 296-307 (2004).
22.
Roberts, M.J., Schirra, H.J., Lavin, M.F. & Gardiner, R.A.
Metabolomics: a novel approach to early and noninvasive prostate
cancer detection. Korean J Urol. 52, 79-89 (2011).
23.
Forster, J., Famili, I., Fu, P., Palsson, B.O. & Nielsen, J. Genomescale reconstruction of the Saccharomyces cerevisiae metabolic
network. Genome Research 13, 244-253 (2003).
24.
Hall, R., et al. Plant metabolomics: The missing link in functional
genomics strategies. . Plant Cell 14, 1437-1440 (2002).
25.
Wishart, D.S., et al. HMDB: the human metabolome database.
Nucleic Acids Res. 35, D521-D526 (2007).
34
26.
Nikiforova, V.J., Daub, C.O., Hesse, H., Willmitzer, L. & Hoefgen, R.
Integrative gene-metabolite network with implemented causality
deciphers informational fluxes of sulphur stress response. J. Exp.
Bot. 56, 1887-1896 (2005).
27.
Kitano, H. Computational systems biology. Nature 420, 206-210
(2002).
28.
Oltvai, Z.N. & Barabasi, A.L. Life's complexity pyramid. Science 298,
763-764 (2002).
29.
Barabasi, A.L. & Oltvai, Z.N. Network biology: Understanding the
cell's functional organization. Nature Reviews Genetics 5, 101-U115
(2004).
30.
Koek, M.M., Jellema, R.H., van der Greef, J., Tas, A.C. &
Hankemeier, T. Quantitative metabolomics based on gas
chromatography mass spectrometry: status and perspectives.
Metabolomics 7, 307-328 (2011).
31.
Fiehn, O., et al. The metabolomics standards initiative (MSI).
Metabolomics 3, 175-178 (2007).
32.
Goodacre, R., et al. Proposed minimum reporting standards for data
analysis in metabolomics. Metabolomics 3, 231-241 (2007).
33.
Lindon, J.C., et al. Summary recommendations for standardization
and reporting of metabolic analyses. Nature Biotechnology 23, 833838 (2005).
34.
WHO. The global burden of disease: 2004 update. (World Health
Organization, Geneva, 2008).
35.
Socialstyrelsen. http://www.socialstyrelsen.se/. (2011).
36.
Zhu, X., et al. Risk-Based Prostate Cancer Screening. Eur Urol
Nov(2011).
37.
Madu, C. & Lu, Y. Novel diagnostic biomarkers for prostate cancer. J
Cancer Oct 6, 150–177 (2010).
38.
Sardana, G., Jung, K., Stephan, C. & Diamandis, E.P. Proteomic
analysis of conditioned media from the PC3, LNCaP, and 22Rv1
prostate cancer cell lines: Discovery and validation of candidate
35
prostate cancer biomarkers. Journal of Proteome Research 7, 33293338 (2008).
39.
Fradet, Y. Biomarkers in prostate cancer diagnosis and prognosis:
beyond prostate-specific antigen. Current Opinion in Urology 19,
243-246 (2009).
40.
Steuber, T., O'Brien, M.F. & Lilja, H. Serum markers for prostate
cancer: A rational approach to the literature. European Urology 54,
31-40 (2008).
41.
Ohlson, N., Wikstrom, P., Stattin, P. & Bergh, A. Cell proliferation
and apoptosis in prostate tumors and adjacent non-malignant
prostate tissue in patients at different time-points after castration
treatment. Prostate 62, 307-315 (2005).
42.
Lissbrant, I.F., Lissbrant, E., Damber, J.E. & Bergh, A. Blood vessels
are regulators of growth, diagnostic markers and therapeutic targets
in prostate cancer. Scandinavian Journal of Urology and
Nephrology 35, 437-452 (2001).
43.
Warburg, O. Origin of cancer cells. Science 123, 309-314 (1956).
44.
Warburg, O. Respiratory impairment in cancer cells. Science 124,
269-270 (1956).
45.
Mazurek, S. & Eigenbrodt, E. The tumor metabolome. Anticancer
Res. 23, 1149-1154 (2003).
46.
Koppenol, W.H., Bounds, P.L. & Dang, C.V. Otto Warburg's
contributions to current concepts of cancer metabolism. Nature
Reviews Cancer 11, 325-337 (2011).
47.
Glunde, K., Jacobs, M.A. & Bhujwalla, Z.M. Choline metabolism in
cancer: implications for diagnosis and therapy. Expert Review of
Molecular Diagnostics 6, 821-829 (2006).
48.
Kim, Y.I. Folate and colorectal cancer: An evidence-based critical
review. Molecular Nutrition & Food Research 51, 267-292 (2007).
49.
Costello, L.C. & Franklin, R.B. Concepts of citrate production and
secretion by prostate. 1.Metabolic relationships. Prostate 18, 25-46
(1991).
36
50.
Costello, L.C. & Franklin, R.B. Novel role of zinc in the regulation of
prostate citrate metabolism and its implications in prostate cancer.
Prostate 35, 285-296 (1998).
51.
Costello, L.C. & Franklin, R.B. 'Why do tumour cells glycolyse?':
From glycolysis through citrate to lipogenesis. Molecular and
Cellular Biochemistry 280, 1-8 (2005).
52.
Vance, D.E. & Vance, J.E. Biochemistry of lipids, lipoproteins, and
membranes (Elsevier, Amsterdam; Boston, 2008).
53.
Swinnen, J.V., et al. Androgens, lipogenesis and prostate cancer.
Journal of Steroid Biochemistry and Molecular Biology 92, 273279 (2004).
54.
Ettinger, S.L., et al. Dysregulation of sterol response elementbinding proteins and downstream effectors in prostate cancer during
progression to androgen independence. Cancer Research 64, 22122221 (2004).
55.
Sreekumar, A., et al. Metabolomic profiles delineate potential role
for sarcosine in prostate cancer progression. Nature 457, 910-914
(2009).
56.
Mor, G., et al. Serum protein markers for early detection of ovarian
cancer. Proc. Natl. Acad. Sci. U. S. A. 102, 7677-7682 (2005).
57.
Chan, E.C.Y., et al. Metabolic Profiling of Human Colorectal Cancer
Using High-Resolution Magic Angle Spinning Nuclear Magnetic
Resonance (HR-MAS NMR) Spectroscopy and Gas Chromatography
Mass Spectrometry (GC/MS). Journal of Proteome Research 8, 352361 (2009).
58.
Denkert, C., et al. Mass spectrometry-based metabolic profiling
reveals different metabolite patterns in invasive ovarian carcinomas
and ovarian borderline tumors. Cancer Research 66, 10795-10804
(2006).
59.
Denkert, C., et al. Metabolite profiling of human colon carcinoma deregulation of TCA cycle and amino acid turnover. Molecular
Cancer 7(2008).
60.
Issaq, H.J., et al. Detection of bladder cancer in human urine by
metabolomic
profiling
using
high
performance
liquid
37
chromatography/mass spectrometry. Journal of Urology 179, 24222426 (2008).
61.
Maeda, J., et al. Possibility of multivariate function composed of
plasma amino acid profiles as a novel screening index for non-small
cell lung cancer: a case control study. Bmc Cancer 10(2010).
62.
Urayama, S., Zou, W., Brooks, K. & Tolstikov, V. Comprehensive
mass spectrometry based metabolic profiling of blood plasma reveals
potent discriminatory classifiers of pancreatic cancer. Rapid
Communications in Mass Spectrometry 24, 613-620 (2010).
63.
Lokhov, P.G., Dashtiev, M.I., Moshkovskii, S.A. & Archakov, A.I.
Metabolite profiling of blood plasma of patients with prostate
cancer. Metabolomics 6, 156-163 (2010).
64.
Thysell, E., et al. Metabolomic Characterization of Human Prostate
Cancer Bone Metastases Reveals Increased Levels of Cholesterol.
Plos One 5(2010).
65.
Lokhov PG, Dashtiev MI, Moshkovskii SA & Archakov, A.I.
Metabolite profiling of blood plasma of patients with PCa.
Metabolomics 6, 156-163 (2009).
66.
Dunn, W.B. & Ellis, D.I. Metabolomics: Current analytical platforms
and methodologies. Trac-Trends in Analytical Chemistry 24, 285294 (2005).
67.
Fiehn, O., et al. Metabolite profiling for plant functional genomics.
Nature Biotechnology 18, 1157-1161 (2000).
68.
Glinski, M. & Weckwerth, W. The role of mass spectrometry in plant
systems biology. Mass Spectrom Rev 25, 173-214 (2006).
69.
Jonsson, P., et al. A strategy for modelling dynamic responses in
metabolic samples characterized by GC/MS. Metabolomics 2, 135143 (2006).
70.
Welthagen, W., et al. Comprehensive two-dimensional gas
chromatography-time-of-flight mass spectrometry (GC x GC-TOF)
for high resolution metabolomics: biomarker discovery on spleen
tissue extracts of obese NZO compared to lean C57BL/6 mice.
Metabolomics 1, 65-73 (2005).
38
71.
Villas-Bôas, S., Mas, S., Akesson, M., Smedsgaard, J. & Nielsen, J.
Mass spectrometry in metabolome analysis. Mass Spectrom Rev 25,
613-646 (2005).
72.
Tauler, R. Multivariate curve resolution applied to second order
data. Chemometr Intell Lab 30, 133-146 (1995).
73.
Jonsson, P., et al. High-throughput data analysis for detecting and
identifying differences between samples in GC/MS-based
metabolomic analyses. Anal Chem 77, 5635-5642 (2005).
74.
Karjalainen, E.J. The Spectrum Reconstruction Problem - Use of
Alternating Regression for Unexpected Spectral Components in 2Dimensional Spectroscopies. Chemometr Intell Lab 7, 31-38 (1989).
75.
Manne, R. & Grande, B.V. Resolution of two-way data from
hyphenated chromatography by means of elementary matrix
transformations. Chemometr Intell Lab 50, 35-46 (2000).
76.
Wold, S. Chemometrics; What do we mean with it, and what do we
want from it? Chemometr Intell Lab 30, 109-115 (1995).
77.
Trygg, J., Holmes, E. & Lundstedt, T. Chemometrics in
metabonomics. Journal of Proteome Research 6, 469-479 (2007).
78.
Holmes, E. & Antti, H. Chemometric contributions to the evolution
of metabonomics: mathematical solutions to characterising and
interpreting complex biological NMR spectra. Analyst 127, 15491557 (2002).
79.
Box, G.E.P., Hunter, W.G. & Hunter, J.S. Statistics for
Experimenters: An Introduction to Design, Data Analysis, and
Model Building., (John Wiley & Sons, Baltimore, 1978).
80.
Lundstedt, T., et al. Experimental design and optimization.
Chemometr Intell Lab 42, 3-40 (1998).
81.
Gullberg, J., Jonsson, P., Nordstrom, A., Sjostrom, M. & Moritz, T.
Design of experiments: an efficient strategy to identify factors
influencing extraction and derivatization of Arabidopsis thaliana
samples in metabolomic studies with gas chromatography/mass
spectrometry. Anal Biochem 331, 283-295 (2004).
82.
A, J., et al. Extraction and GC/MS analysis of the human blood
plasma metabolome. Anal Chem 77, 8086-8094 (2005).
39
83.
Antti, H., et al. Statistical experimental design and partial least
squares regression analysis of biofluid metabonomic NMR and
clinical chemistry data for screening of adverse drug effects.
Chemometr Intell Lab 73, 139-149 (2004).
84.
Ransohoff, D.F. Opinion - Bias as a threat to the validity of cancer
molecular-marker research. Nature Reviews Cancer 5, 142-149
(2005).
85.
Wold, S., Esbensen, K. & Geladi, P. Principal Component Analysis.
Chemometr Intell Lab 2, 37-52 (1987).
86.
Jackson, J.E. A User's Guide to Principal Components, (John Wiley
& Sons, New York, 1991).
87.
Wold, S., Sjostrom, M. & Eriksson, L. PLS-regression: a basic tool of
chemometrics. Chemometr Intell Lab 58, 109-130 (2001).
88.
Hoskuldsson, A. A Combined Theory for Pca and Pls. J Chemometr
9, 91-123 (1995).
89.
Pohjanen, E., et al. A Multivariate Screening Strategy for
Investigating Metabolic Effects of Strenuous Physical Exercise in
Human Serum. Journal of Proteome Research 6, 2113 -2120 (2007).
90.
Wuolikainen, A., Moritz, T., Marklund, S.L., Antti, H. & Andersen,
P.M. Disease-Related Changes in the Cerebrospinal Fluid
Metabolome in Amyotrophic Lateral Sclerosis Detected by
GC/TOFMS. Plos One 6(2011).
91.
Wibom, C., et al. Metabolomic Patterns in Glioblastoma and
Changes during Radiotherapy: A Clinical Microdialysis Study.
Journal of Proteome Research 9, 2909-2919 (2010).
92.
Stenlund, H., et al. Monitoring kidney-transplant patients using
metabolomics and dynamic modeling. Chemometr Intell Lab 98,
45-50 (2009).
93.
Jonsson, P., et al. A strategy for identifying differences in large series
of metabolomic samples analyzed by GC/MS. Anal Chem 76, 17381745 (2004).
94.
Griffin, J.L., et al. An integrated reverse functional genomic and
metabolic approach to understanding orotic acid-induced fatty liver.
Physiological Genomics 17, 140-149 (2004).
40
95.
Griffin, J.L. & Bollard, M.E. Metabonomics: Its potential as a tool in
toxicology for safety assessment and data integration. Current Drug
Metabolism 5, 389-398 (2004).
96.
Nicholson, J.K., Connelly, J., Lindon, J.C. & Holmes, E.
Metabonomics: a platform for studying drug toxicity and gene
function. Nat Rev Drug Discov 1, 153-161 (2002).
97.
Clarke, C.J. & Haselden, J.N. Metabolic Profiling as a Tool for
Understanding Mechanisms of Toxicity. Toxicologic Pathology 36,
140-147 (2008).
98.
Jobard, E., Tredan, O., Elena, B., Blaise, B.J. & Bachelot, T.
Metabolomics: a novel tool for translational research in oncology.
Oncologie 12, 409-415 (2010).
99.
Martens, H. & Naes, T. Multivariate calibration, (John Wiley &
Sons, Chichester, 1989).
100.
Trygg, J. & Wold, S. Orthogonal projections to latent structures (OPLS). J Chemometr 16, 119-128 (2002).
101.
Cloarec, O., et al. Statistical total correlation spectroscopy: An
exploratory approach for latent biomarker identification from
metabolic H-1 NMR data sets. Anal Chem 77, 1282-1289 (2005).
102.
Wiklund, S., et al. Visualization of GC/TOF-MS-based metabolomics
data for identification of biochemically interesting compounds using
OPLS class models. Anal Chem 80, 115-122 (2008).
103.
Bylesjo, M., et al. OPLS discriminant analysis: combining the
strengths of PLS-DA and SIMCA classification. J Chemometr 20,
341-351 (2006).
104.
Eriksson, L., Johansson, E., Kettaneh-Wold, N. & Wold, S. Multiand Megavariate Data Analysis Principles and Applications,
(Umetrics Academy, Umeå, 2001).
105.
Pohjanen, E., et al. A multivariate screening strategy for
investigating metabolic effects of strenuous physical exercise in
human serum. Journal of Proteome Research 6, 2113-2120 (2007).
106.
Eriksson, L., Johansson, E., Muller, M. & Wold, S. Cluster-based
design in environmental QSAR. Quantitative Structure-Activity
Relationships 16, 383-390 (1997).
41
107.
Jonsson, P., et al. Predictive metabolite profiling applying
hierarchical multivariate curve resolution to GC-MS datas - A
potential tool for multi-parametric diagnosis. Journal of Proteome
Research 5, 1407-1414 (2006).
108.
Pohjanen, E., et al. Statistical multivariate metabolite profiling for
aiding biomarker pattern detection and mechanistic interpretations
in GC/MS based metabolomics. Metabolomics 2, 257-268 (2006).
109.
Chorell, E., Moritz, T., Branth, S., Antti, H. & Svensson, M.B.
Predictive Metabolomics Evaluation of Nutrition-Modulated
Metabolic Stress Responses in Human Blood Serum During the
Early Recovery Phase of Strenuous Physical Exercise. Journal of
Proteome Research 8, 2966-2977 (2009).
110.
Wuolikainen, A., Andersen, M., Moritz, T., Marklund, S. & H, A. ALS
patients with mutations in the SOD1 gene have an unique
metabolomic profile in the cerebrospinalfluid compared with ALS
patients without mutations. Molecular Genetics and metabolism
accepted(2011).
111.
Marengo, E. & Todeschini, R. A New Algorithm for Optimal,
Distance-Based Experimental-Design. Chemometr Intell Lab 16, 3744 (1992).
112.
Crnalic, S., et al. Nuclear androgen receptor staining in bone
metastases is related to a poor outcome in prostate cancer patients.
Endocrine-Related Cancer 17, 885-895 (2010).
113.
Freeman, M.R. & Solomon, K.R. Cholesterol and benign prostate
disease. Differentiation 82, 244-252 (2011).
114.
Awad, A.B., Fink, C.S., Williams, H. & Kim, U. In vitro and in vivo
(SCID mice) effects of phytosterols on the growth and dissemination
of human prostate cancer PC-3 cells. European Journal of Cancer
Prevention 10, 507-513 (2001).
115.
Zhuang, L., Kim, J., Adam, R., Solomon, K. & MR., F. Cholesterol
targeting alters lipid raft composition and cell survival in prostate
cancer cells and xenografts. J Clin Invest. 115, 959-968 (2005).
116.
Zhuang, L.Y., Lin, J.Q., Lu, M.L., Solomon, K.R. & Freeman, M.R.
Cholesterol-rich lipid rafts mediate Akt-regulated survival in
prostate cancer cells. Cancer Research 62, 2227-2231 (2002).
42
117.
Shaffner, C. Prostatic cholesterol metabolism: regulation and
alteration. Prog Clin Biol Res. 75A, 279-324 (1981).
118.
Nordstrand, A., et al. Establishment and validation of an in vitro coculture model to study the interactions between bone and prostate
cancer cells. Clinical & Experimental Metastasis 26, 945-953
(2009).
119.
Nordstrand, A., et al. Combined simvastatin and anti-IGF1R therapy
of prostate cancer cells in co-culture with bone. Manuscript in
preparation (2012).
120.
Dillard, P.R., Lin, M.F. & Khan, S.A. Androgen-independent prostate
cancer cells acquire the complete steroidogenic potential of
synthesizing testosterone from cholesterol. Molecular and Cellular
Endocrinology 295, 115-120 (2008).
121.
Montgomery, R.B., et al. Maintenance of intratumoral androgens in
metastatic prostate cancer: A mechanism for castration-resistant
tumor growth. Cancer Research 68, 4447-4454 (2008).
43
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising