I would like to thank the organizers for giving me this opportunity to talk to you today.
I want to talk about the identification of biomarkers for obesity using microarrays and metabonomics.
This quote from Bill Lasley summarizes the value of metabolomics or metabonomics.
I will briefly talk about growth hormone and its action and the animal model we used in the study. I will then focus on microarray data analysis and metabonomic confirmations. Finally I will talk about the potential biomarkers identified and how they can be used in clinical studies.
It acts through its receptor on the cell membrane. Binding of the hormone induces a number of intracellular signalling pathways, such as STATs, MAPK, PI3K and PKC. This signalling leads to activation of gene expression. Their relative contributions of the different pathways to various actions of GH are still unclear.
In order to study the gene regulation induced by GH we have used GH receptor mutant mice, which I will talk about in the moment. For this study we have used livers from male adult mice. The samples were hybridized to Affymetrix U74v2 arrays. For the metabonomic study we have used NMR spectroscopy to identify the changes in metabolic profiles in these mice. We collected urine from fasted animals aged betwween 2 and 12 months. Finally, the biomarkers derived from this study are currently being assessed in human clinical screening.
The animal model we used are mice with mutations in the intracellular domain of the growth hormone receptor. This has led to changes in STAT5 signaling pathways. However, a 70% reduction of STAT5 levels in response to GH has produced surprisingly little change at the transcriptional level – only 35 transcripts were changed, in comparison to 121 when all of the STAT5 is absent. What are the relationships between the genes that are differentially expressed?
We used a multivariate analysis tool - GeneRaVE to determine how the different transcripts interact with each other. There were some links between the genes but overall, majority of trasncripts pointed towards the response. This was not surprising since GHR is a membrane protein and elicits a metabolic response in the body. GeneRaVE also identified some biomarkers that can differentiate between the 4 groups of animals, WT, mutant 569, 391 and GHR KO.
One of the main aims of this study was to identify the metabolic outcomes of the changed GH signaling. Using pathway analysis tools and gene Ontology we have identified a number of metabolic pathways, which were changed. Important alterations were found in glutathione, androgen/steroid metabolism, glycolysis/gluconeogenesis, TCA cycle and fatty acid metabolism. Green are genes down-regulated and orange to red are up-regulated.
A number of genes have been identified as good markers separating the mouse groups, with those 3 being the main markers, and 19 others were selected in the subsequent process. There are a few problems with these markers. While they might be useful in the mouse model they really bring not much to our understanding of the effects a lack or alteration of GH signaling has on physiology. GH-deficiency becomes pronounced with age and one of the hallmarks is increased central obesity. Why does that happen? With these results we can possibly have a few ideas and try to see if these genes are increased or decreased in people but really that is not the most suitable way forward. We need a different tool for doing it.
Metabonomic analysis has been around for some 15 years. What is metabonomics? It is a way of identification of metabolites in a complex mixture using statistical tools. It can be a quantitative method. A number of various metabonomic platforms exist. We have decide to use NMR as it is non-destructive, and requires very little sample preparation. As we all know any alterations in physiology due to mutations, drugs, diet produce metabolic changes that can be measured. NMR platform uses the information in 1D spectrum to compare different individuals and separate them on the basis of similarity in their metabolic profiles. Thus it uses the entire dataset to really enable differentiation. Once the groups are established the metabolites that contribute the most to the differentiation can be identified from the loadings plot. PCA (finding the directions of largest variation) (finding the directions of largest variation)
PCA analysis is good to use in such experiments as it is unsupervised method and therefore will not be biased by external information. It was able to separate the samples of WT from mutant 391, while the mutant 569 was a bit less obviously separated, appearing to have an intermediate phenotype. The PLS analysis has made the separation much better but still the WT and mutant 391 remained well separated while the mutant 569 showed still the intermediate phenotype.
The observations from the arrays and metabonomic analysis have been supported further by phenotypic characterization of the mice. Their appearance clearly changed over the 10 months and the gain in weight, especially in mutant 569 (POINT) was due to an increase in the amount of the fat tissue. It is also clear from this data that mutant 569 is initially very similar to the WT but around 6 months of age it starts to resemble the mutant 391. Obesity is often linked to insulin resistance and diabetes. In addition to looking at the urine using metabonomics we also did standard blood analysis.
This has been done at ages from 4 to 13 months. Here are the results from 10 and 13 months. Levels of glucose in mutant 569 have reached the pre-diabetic levels. Lipid markers such as HDLs and FFAs were easily measured in blood and showed increase in mutant 391. Now an obvious question is what did all this give us? How to figure out where all those puzzle pieces fit?
Here are the results from the microarrays. A couple of clusters of genes, some identified by GO some by hand annotation as belonging to a particular group. The genes in red are up-regulated and the ones in blue down-regulated. The up-regulation of TCA and beta-oxidation genes was impossible to explain. Subsequently we measured metabolite levels. TCA cycle intermediates fitted with the transcripts but were all decreased. The most likely explanation is compensation for low energy production. Finally, we put all pieces together to generate this image. We can now explain the increase in creatine and TMA/TMAO/DMA levels. Cysteine is being shunted towards glycine rather than taurine overflowing the pathway and increasing not only creatine production but also TMA/ TMAO and DMA. By using microarays and metabonomics we were able to identify the biomarkers which can now be validated in clinical studies.
However, the main question still remains if any single metabolite or even a couple will be adequate tool in large clinical studies. Most likely the metabolic fingerprints are going to be more useful, where some metabolites are increased while others are reduced and the changes will occur in only certain metabolic pathways, possibly characteristic of a given disease. In a way this is not unlike the microarray profiling of for example, cancer samples, where a set of markers is identified and this classifier set is then used to predict the behaviour or phenotype of unknown sample.
We are currently assessing the markers identified in the mouse study in clinical cohorts.
We have a number of patient cohorts, from which various types of samples have been collected. While NMR is our primary platform we will also use Mass Spectroscopy to separate more complex regions of the spectrum into smaller fragments. Biomakers we have are being tested for validity in these studies, we also are generating the metabolic fingerprints of human populations and establishing the health/disease trajectories. As these studies have not been concluded yet, I cannot comment on the confirmation of markers. Our study has shown that profiles can change overtime and mutant 569 in particular is a great model for studying the development of obesity and diabetes.
I would like to thank a Horst Schirra who was instrumental in development of the metabonomics method, Cameron Anderson for performing analysis, Linda for great help in the stinky collection process, Bill Wilson from CSIRO for GeneRaVE analysis of the data. Mike Waters for believing that this project is worth pursuing. Current clinical studies are performed in collaboration with Mater hospital metabolic disease unit and in part form a project of my student Shaffinaz. Thank you
Identifying metabolic markers of pre-obesity regulated by growth hormone using microarrays, confirmation through metabonomics and use in clinical applications
IMB, UQ, Brisbane
Dr A. M. Lichanska, UQ, Brisbane 2
Identifying metabolic markers of pre-
obesity regulated by growth hormone
using microarrays, confirmation
through metabonomics and use in
Dr Agnieszka M. Lichanska, PhD
Oral Biology and Pathology, School of Dentistry and IMB,
University of Queensland,
Dr A. M. Lichanska, UQ, Brisbane 3
Genomics and proteomics tells you what
might happen, but metabolomics tells you
what actually did happen.
- Bill Lasley, University of California, Davis
Dr A. M. Lichanska, UQ, Brisbane 4
Overview of the presentation
• Animal model
• Array data - GO and metabolic pathway
• Metabonomics platform
• Data analysis and biomarker identification
• Analysis of the metabolism
• Clinical assessment of the biomarkers
Dr A. M. Lichanska, UQ, Brisbane 5
Roles of GH
Growth Hormone (GH) is the major
regulator of somatic growth and
Dr A. M. Lichanska, UQ, Brisbane 10
Array data - metabolic pathway analysis
TCA cycle Fatty acid
Pathway Miner – www.biorag.org
Dr A. M. Lichanska, UQ, Brisbane 11
Potential biomarkers from the
• Es31 - esterase 31
• Hsd3b5 – hydroxysteroid dehydrogenase
• RCK/p54 – RNA helicase
• 19 others: Csad, Igf1, Egfr, Keg1, Socs2, Fabp5,
Fos, Fmo3, Dbp, Rgs16, Hao3, H19, Igfbp1, Cyp2b9,
Mt1, Sult2a2, Ghr, 1100001G20Rik
How do these markers help us in
understanding the physiology of GH action?
Dr A. M. Lichanska, UQ, Brisbane 12
Metabonomic analysis using NMR platform
Changes in metabolite
NMR PCA/PLS-DA analysis
H2O and urea excluded
Bucket table (bucket width 0.05ppm)
Dr A. M. Lichanska, UQ, Brisbane 13
PCA and PLS-DA analysis
Dr A. M. Lichanska, UQ, Brisbane 19
TMA, TMAO, DMA
Is it the individual biomarkers or rather the
metabolic fingerprint that is going to be more
useful for diagnostics?
Fatty acid metabolism
Dr A. M. Lichanska, UQ, Brisbane 20
Assessment of biomarkers in clinical
samples – taking results from mouse model
into the clinic
• GH-deficient patients
• GH/GHR mutation, GH - insensitivity
• Prader-Willi syndrome children
• Follow up of GH replacement therapy
• Identification of a type of GH-deficiency
• Identification of pre-obesity state
Dr A. M. Lichanska, UQ, Brisbane 21
Assessment of biomarkers in clinical
samples - testing clinical samples
Urine Saliva Blood
Dr A. M. Lichanska, UQ, Brisbane 22
• Microarray analysis has identified the metabolic
processes affected by mutations in GHR and some
• Metabonomic analysis has confirmed the predicted
• The small changes in microarrays lead to
significant alterations in metabolism
• Both methods identified the intermediate
phenotype represented by the mutant 569
• Metabolic profiling of human samples will focus
first on the metabolites identified in mouse study.
Dr A. M. Lichanska, UQ, Brisbane 23
Take home message
• The markers identified in microarray studies have
been useful for metabolic differentiation of
individual groups using metabonomic analysis.
• Pre-obesity syndrome was identified using both
microarrays and post-array NMR analysis.
• These results form basis of human clinical
analysis using metabonomics.
Dr A. M. Lichanska, UQ, Brisbane 24
Horst J. Schirra
David J. Craik
Mike J. Waters
School of Dentistry/IMB
Shaffinaz Abd Rahman
University of Ohio