Download Accepted version

Serum Metabolomics in Animal Models and Human Disease
Abstract
Purpose of review: To highlight some of the ways in which serum metabolomics has been
used in a range of recent human and animal studies. The main themes are the importance of
understanding the underlying variation in human metabolism and the use of serum
metabolomics in disease profiling.
Recent findings: Several recent studies have attempted to use serum metabolomics to
develop non-invasive biomarkers of disease and/or track the consequences of nutritional
and genetic interventions. Many advances have been made with common changes being
identified in ageing, the menopause and cancer but several problems of interpretation have
emerged from these studies. These include the small sample sizes in most human studies
and the differences between human and rodent metabolomes. However, a large scale
metabolic screen of over 1000 ‘healthy’ humans (the humsermet project) has highlighted
many variables that may be used to refine the interpretation and design of previous and
future human studies alike, in addition to researcher data mining.
Summary: Several common serum metabolome alterations have now been identified but
many inconsistencies remain.
The recent construction of a human serum metabolome
database should help resolve this issue and be informative in the design of future human
and animal models studies.
Key words: metabolomics, cancer, serum, study design
1.) Serum Metabolomics
The metabolome; the collective term for all of the intermediates and waste products of the
biochemical reactions that occur within a biological sample, allows us to derive insight into
the metabolic processes that underpin specific diseases, accurately quantify nutrient or toxin
exposure, and link genetic makeup to molecular mechanisms in drug and nutrient
metabolism. The metabolome is increasingly being recognised as a rich source of holistic
information about the health of an organism, as evidenced by the large increase in
metabolomics studies published in the last 4 years (fig 1).
Figure 1: Number of serum metabolomics papers published in the last four years, as shown
on PubMed at the time of writing (May 2015)
Although it is possible to collect metabolomics data from many different compartments (for
example blood, saliva, urine, soft tissue, and conditioned cell culture media), blood plasma is
clinically particularly useful for a number of reasons: 1. Blood is routinely taken in many
clinical studies, so obtaining serum is usually not a challenge 2. Serum contains molecules
that represent ‘short term’ biomarkers, so is reflective of the current state of an organism –
this can be useful for tracking metabolism over a set time course in animal models or
humans 3. Serum is amenable to many analysis platforms including several chromatography
coupled to Mass Spectrometry (MS) arrangements, Nuclear Magnetic Resonance (NMR)
spectrometry, Enzyme Linked Immunosorbant Assays (ELISAs) and enzymatic assays 4.
Samples can be stably stored in a freezer, and only small sample volumes are required for
all of the methods listed above.
In the search for novel biomarkers and insights into biological processes a common
approach to employ is an untargeted screen to detect as many molecules as possible,
coupled with multivariate analysis such as principal components analysis (PCA) to make a
comparison between a control and test group(s) without having prior knowledge as to which
molecules to compare. Candidate molecules identified using databases such as the Human
Metabolome Database (HMDB) can then be further assessed quantitatively using labelled
standards in what is known as targeted metabolomics. (figure 2 –see also review on
techniques or astarita and Langridge 2013)
Figure 2: General metabolomics workflow (MS = mass spectrometry NMR = nuclear
magnetic resonance spectrometry)
2.) Use in animal models
Animal models offer the ability to manipulate genes of interest, or induce pathology in a
controlled fashion and investigate the metabolic implications on a whole organism level.
(a) Ageing
Wilson and co workers (Wilson Cell Death and Disease 2015) studied the caspase 2 null
mouse that amongst other things displays premature ageing. In common with aged wild type
mice, young caspase 2 null mice showed an increase in several saturated fatty acids and a
lower abundance of glucose and mannose-6-phosphate relative to their wild type
counterparts. However, most of the serum metabolites they found significantly associated
with ageing were quite different from those found in human ageing (Menni et al 2013) and in
the conditioned medium of cultured senescent human cells (James et al J. Proteome Res
2015) the omega-3 polyunsaturated fatty acid (eicosapentaenoic acid, EPA; C20:5n3) was
elevated in all 3 studies and these common findings are encouraging.
(b) Cancer LaConti et al have recently published a study identifying a serum metabolic
signature of 50 ions detected by UPLC TOF MS that distinguishes between precancerous
lesions, advanced pancreatic cancer and normal pancreas in a well-established mouse
model, although early stage lesions were less reliably detected. (ref) One of the metabolites
identified, citrate, has also been reported to be elevated in another serum study of
pancreatic cancer in rats, and in the current study the authors were able to ascertain that this
rise in citrate was due to increased citrate synthase expression. However, despite also
being able to separate disease from healthy control based on serum metabolites either with
MS or 1H NMR, older studies of humans with pancreatic cancer using similar techniques do
not show the same specific elevation in citrate. ( No recent ones but Koyabashi et al 2013
seemed to do a good job uramaya 2010).
As well as identifying biomarkers, metabolomics can be used to monitor the effectiveness
and mechanisms involved in potential treatments. Using ovariectomised mice as a model of
post-menopausal osteoporosis, Chen et al investigated the effectiveness of increasing
vitamin D2 levels and maintaining bone density by feeding mice mushrooms that had been
irradiated with UV to increase their vitamin D content. Using NMR followed by spectral
deconvolution and peak identification in Chenomx to analyse the serum of sham,
ovariectomised, ovariectomised + non-irradiated mushroom and ovariectomised + irradiated
mushroom, Chen et al demonstrated that the metabolic profile of the ovariectomised mice on
the irradiated mushroom diet was distinct from that of the other groups. Using PCA they
were able to see that this separation was caused mostly by amino acids and energy
metabolites previously shown to be associated with the bone forming cells osteoblasts. The
mice on the irradiated mushroom diet also showed increased levels of osteocalcin in their
serum, a molecule secreted by osteoblasts, and reduced levels of PYD which is a product of
collagen breakdown associated with the bone resorbing cells osteoclasts. This suggests a
direct link between the level of vitamin D2 and amount of bone maintenance via inhibition of
collagen breakdown and promotion of osteoblasts (ref).
3.) The human metabolome
Equating mechanisms and biomarkers derived from highly controlled experiments in rats or
mice to what we observe in humans, who aside from being a different species are also
heterogeneous in regards to the plethora of factors that affect serum metabolites levels, is a
challenge. In the two biomarker studies mentioned above, the markers found in the animal
models did not all show up in human studies. Is this because the animal models are not a
robust representation of the conditions in humans? After all the metabolites are the footprint
of the biological process, therefore if the same signature is not present between species it
might indicate there are different mechanisms at play. Alternatively, is it because the human
studies simply don’t have enough power, given the variability in the sample, to detect the
signature of the pathology over the background of other confounding factors? Recent work
by Dunn et al (ref) has begun to address this issue:
The
Husermet
project
(http://www.husermet.org/)
which
has
applied
non-targeted
chromatography-mass spectrometry (GC–MS, UPLC–MS + and UPLC–MS - ), Dunn et al
(Metabolomics 2015) to study the hydrophilic and lipophilic metabolic complement of serum
samples in 1200 healthy UK subjects, where healthy is described as having no known
disease at the time. This large study reported that the variation in serum metabolites ranged
from less than 5% to more than 200%due to differences in gender, age, BMI, blood
pressure, and smoking although the variation in the serum metabolome was less than in the
urine metabolome published previously, some of which could be linked to lifestyle or drug
use. For example, highly variable metabolites such as caffeine could be linked to its variable
consumption, N-methylpyrrolidinine may be linked to its use in drug vehicles, salycilic acid
possibly linked to aspirin use or smoking, trehalose could be linked to food additives and
oxidised longer chain fatty acids, acyl carnitines or and two -glutamyl dipeptides (isoleucine
and leucine) could be linked to variations in oxidative stress.
There were several interesting differences in gender including caffeine and 2-aminomalonic
acid being higher in females perhaps indicative of higher coffee and chocolate consumption
and also creatinine and phosphate, indicative of increased muscle breakdown.
Several metabolites also increased with age including citrate, which we have recently
reported to accumulate in the extracellular environment of DNA-damaged and senescent
cells (James et al 2015) and several others only previously reported to correlate with
increasing age in females (Menni et al 2013).
Increased body mass index (BMI) is correlated to increases in body fat, greater risk of insulin
resistance and metabolic disorders including diabetes and cardiovascular diseases and is a
surrogate measure of body fatness. A range of amino acids varied with BMI and short chain
organic
acids
such
as
acetate,
certain
diacylglycerides,
sphingolipids,
lyso-
glycerophospholipids and fatty acids showed a decrease in concentration with increased
BMI and other metabolites showed a female-specific decrease and glycerol-3-phosphate a
male-specific decrease.
Elevated blood pressure (BP) is associated with an increased risk of cardiovascular disease
(CVD). Methionine disulphide decreased with increased BP whereas methionine increased
and although counterintuitive in correlating inversely with reactive oxygen species (ROS), it
might indicate an accumulation of ROS that are not detoxified by methionine. Alternatively
increased methionine might indicate increased activity of the thioredoxin system and/or Sadenosylmethione cycle to regenerate it. Amongst many other correlations were increases in
citrulline and lactate, as well as correlations with urate, triacylglycerides, dipeptides,
glycerophosphocholines and 4-hydroxyphenyllactic acid.
Smoking has been identified as an important risk factor for cancer and CVD. Several amino
acids (aspartate, histidine and lysine), glycerol-3-phosphate and a number of fatty acids,
citrate, lactate biotin and inositol all decreased in smokers when compared with nonsmokers and even ex-smokers.
Arguably as important as the actual observations, are the calculations made by Dunn et al
using this large data set to build a model capable of classifying additional data accurately.
The authors concluded that in their particular study, comparing all ‘healthy’ individuals, 600
samples would have been enough to accurately build a predictive model from the data,
however in studies comparing a healthy group with a control group where variability would
likely be increased the ideal sample number would be higher.
All of these observations, made in a large cohort of healthy individuals, provide an invaluable
reference point for other publications and future work in which a more targeted approach has
been taken, or when the study contains a much smaller number of subjects. Importantly, the
authors explain how they were able to correct for intra-run drift in spectra, as is inevitable
when running such large numbers of samples on different days. Further expansion of these
observations in a robust and reproducible manner, such as described by Dunn et al, is vital
to maximise the usefulness of targeted studies in smaller groups of patients. The use of big
data open access repositories such as MetaboLights and a project to map the metabolome
similar to the Human Genome Project, which has proved so useful in understanding genetic
variation, will greatly enhance the interpretation of serum metabolomics studies.
4. Clinical Testing
Although there are many potential challenges to consider when analysing data, the relatively
non-invasive nature of serum metabolomics has revolutionised clinical studies, allowing
researchers to study biochemical processes in humans like never before. The ability to take
multiple samples over a short space of time has led to informative studies on the temporal
patterns in metabolism of drugs and foods in humans.
(a) Nutrition and Obesity
One recent example is the work of Liu et al studying the metabolic signature associated with
insulin resistance in the serum of healthy versus obese young men at several time points
during an oral glucose tolerance test. Using targeted gas chromatography (GC) MS and
ultra performance liquid chromatography (UPLC) MS they confirmed and expanded upon
previous studies showing that there are significant differences between fasting levels of free
fatty acids in the serum of obese and non-obese men on similar diets, and that levels of
branched chain amino acids and palmitic acid in serum after a meal correlate with insulin
resistance.
(ref) Further investigations into this correlation may lead more effective
interventions to prevent deterioration in the metabolic condition of obese patients.
(b) Aging, CVD and the menopasue
In a very large study of over 10,000 females Auro et al (Nature Communications 2014)
reported that menopause status associated with amino acids increased glutamine, tyrosine
and isoleucine, along with serum cholesterol and atherogenic lipoproteins and additionally in
a subset of women with with glycine and total, monounsaturated, and omega-7 and -9 fatty
acids. The authors concluded that the increase in these amino acids in addition to certain
lipids was related to the menopause in women and might impact in numerous pathways
associated with diabetes and CVDs. Interestingly, the authors also noted age-related
increases in omega-3 polyunsaturated fatty acids and citrate both of which are independent
markers of human ageing and cellular senescence (Menni et al 2013; James et al 2015).
(c) Cancer
Several recent studies have used serum metabolomics as a first step towards developing
more accurate non-invasive tests for cancer and although there has been limited consensus
in the cancer signatures obtained some of this might subsequently be rectified by taking into
account confounding factors such as age, BMI, gender and smoking history as well as the
different platforms used. However, the study by Kumar et al did confirm earlier reports that
sarcosine when assessed by 1H NMR as well as three new metabolites could well help
refine the notoriously unreliable PSA test for prostate cancer. Also two studies (Zamani et al
2015, Zhu et al 2014) identified alterations in bile acid metabolism as being indicative of
colon cancer and one study (Zhu et al 2014) suggested that this pathway may be able to
distinguish malignancies from pre-malignancies. Elevated levels of alanine and glycine have
now been reported in both prostate cancer (Kumar et al 2015) and colon cancer (Zhu et al
2014) and elevated alanine has also been reported previously in the saliva of oral premalignancies, suggesting that this amino acid could be a useful non-invasive marker of
neoplasia or cancer although much further work is needed to verify this as it was not
identified as a biomarker of oral pre-malignancy in the recent larger study of Gupta et al
(2015). Other common features of cancer included alterations in glutamine (liver cancer and
oral cancer).
The recent results are summarised in Table 1.
Sensitivity/
Study
Cancer Type
Reference
Platform
specificity
Metabolite Signature
Size
(%)
Prostate
Kumar et al
1
102
84.4/92.9
Sarcosine
1
70
92.5/93.3
Alanine, glycine pyruvate
H NMR
versus normal
JPR2015
Prostate high
Kumar et al
grade
versus
H NMR
JPR2015
low grade
114
Prostate
Zang et al
fatty acids, amino acids,
UPLCAge
versus normal
JPR 2015
92.1/94.3
MS/MS
lysophospholipids
matched
Inverse relationship:
alpha-ketoglutarate, citrate,
Prostate high
UPLC-
300
MS/MS
Age
GC/MS
matched
Mondul et al
grade
versus
inositol-1-phosphate,
ND
IJC 2015
several
normal
glycerophospholipids
and
fatty acids
Cross et al UPLCColon versus
> 205 per
Cancer
MS/MS
normal
NA
None
group
2014
Zamani
GC/MS
Bile
et
al (Biochem
vitamin
Colon versus
33
Research
normal
acid
1
H NMR
B6
biosynthesis,
metabolism,
plus
ND
methane metabolism, and
control
Internationa
glutathione
metabolism.
l 2015)
Elevated glycine
histidine,
glycocholate,
hippuric acid, malonic
acid/3-hydroxybutyrate
(3HBA),
glycochenodeoxycholate,
Colon versus Zhu
et
al UPLC158
normal
JPR2014
ND
leucic
acid,
methionine,
MS/MS
maleic acid, linolenic acid,
hydroxyproline,
2-
aminoadipate,
N-
acetylglycine, and
glyceraldehyde.
adenosine,
alanine,
phosphoenolpyruvate
(PEP),
glyceraldehyde,
glycocholate,
hippuric
acid,
Colon versus Zhu
et
glycochenodeoxycholate,
al UPLC132
pre-malignant
JPR2014
MS/MS
ND
trimethylamine-N-oxide, Nacetylglycine,
hydroxyproline/
aminolevulinate,
dimethylglycine,
linolenic
acid, leucic
acid, and pantothenate
Chen et al
Biochem
Lung
versus
Sphingosine+,
GC/MS
Research
60
post-operative
96.7/90.0+
LC/MS
serine
Internationa
l 2014
Sphingosine,
Chen et al
glycerophospho-N-
Biochem
Lung
arachidonoyl
versus
GC/MS
Research
60
normal
96.7/90.0+
ethanolamine+,
linolenic
LC/MS
Internationa
acid,
l 2014
9,12-octadecadienoic acid,
oleic acid, serine
Liver
(even
Depletion
Zeng et J. TOF/MS,
of
glutamine,
small tumours)
98
98.0/82.0+
tryptophan+. Accumulation
Proteome
electrospr
Res 2014
ay
of 2-hydroxybutyric acid
Targeted
Exploratory study
versus normal
and cirrhosis
Breast
Frail
Corona et al
versus J.
Cell
tandem
unhealthy
Physiol
versus healthy
2014
89
ND
MS
Oral
versus Bag
et
al
1
H
lipids
and
trimethyl
30
normal
BBRC
Several amino acids and
amine
N-oxide,
ND
13
C NMR
malonate
Gupta et al
Oral
versus Clinica
glutamine,
1
H NMR
normal
Chimica
Acta 2015
175
propionate,
90.0/94.0
acetone, and choline
Gupta et al
Oral
versus Clinica
glutamine,
1
H NMR
pre-malignant
200
acetone,
90.0/92.0
Chimica
acetate, choline
Acta 2015
Table 1: Recent studies of cancer, Where pathways appear more than once they are
indicated in bold and where specificity and sensitivity values were given for individual
metabolites the best values are given and indicated by +.
In conclusion, serum metabolomics has already begun to give us better insight into the
complex effects of pathology and treatment on whole biological systems.
The biggest
challenges are in getting suitable numbers of participants, collecting relevant metadata on
factors known to impact on metabolite profiles and analysis of the data which takes into
account those known factors in order to derive meaningful information about the condition of
interest. So far the any consensus between the serum metabolomes of human conditions
and with animal models designed to investigate them has been limited although the
publication of the Husermet project (Dunn et al 2015) should perhaps help standardise these
studies in the future. in the understanding of the mechanisms underpinning the human
metabolome.
Interaction
between
clinicians,
biologists,
systems
biologists
bioinformaticians is essential for this discipline to fulfil its potential as the last omics.
and