I had the pleasure of presenting a 10 minute oral abstract in the Science session at Rheumatology 2013 on Thursday 26th April. Below is a quasi-transcript of the talk annotated with presented data. There are minor edits over points where I would have been waving a pointer at the screen and relevant slide-elements have been extracted as single figures for clarity. The original Powerpoint file is also available, as usual under CC-BY-SA.
I’ve left the jokes in. If you are in any doubt, they were better in person.
Metabolomics
Metabolomics, is the “study of the unique fingerprints that result from a combination of cellular processes, and the environment”. It’s often represented as shown, as the bottom rung on a hierarchy leading from the genome, through transcribed genes and translated proteins to the product of their reactions.
Metabolomics may be untargeted or targeted, the former being analysis of the whole metabolome of a given system, the latter focusing on a particular metabolic process.
In vitro metabolomic studies have been used successfully in cancer to identify key metabolic pathways targets for drugs – such as 2-deoxyglucose against the glycolytic pathway. Meanwhile, analysis of patient biofluids has shown diagnostic and prognostic potential in many inflammatory diseases. For example, our group has recently shown the power of urine metabolites in predicting response to anti-TNFa therapy in RA.
What we have sought to do in this study therefore, is apply the in vitro metabolomics techniques successfuly used in cancer research, to a model of inflammatory cells in the rheumatoid joint.
Cellular model: Macrophages
Macrophages are innate immune cells, with inflammatory and homeostatic roles. During inflammation infiltrating macrophages respond to local stimuli to differentiate into pro-inflammatory or anti-inflammatory phenotypes.
M1 or ‘classically activated’ macrophages are pro-inflammatory, releasing IL-6, IL-1;
While M2 ‘alternatively activated’ macrophages are a broader classification involved in resolution, repair, and releasing IL-10 among cytokines.
Macrophages are one of the key cells in the rheumatoid synovium. They are abundant and activated in disease, and correlated with both severity and bone destruction.
Environment: Hypoxia
Healthly synovium is a hypoxic environment with normal tissues held at ~8% O2. However, perfusion decreases with disease and negatively correlates with disease severity (as shown by Ng et al).
Raised synovial pressure associated with inflammation also drives intermittent vessel occlusion and local reperfusion injury.
Interestingly for us, macrophages have been shown to accumulate in such hypoxic tissues.
So using metabolomic tools, we have attempted to address the question: Does the hypoxic synovial environment affect macrophage differentiation, function and metabolism?
Experimental design
For this series of experiments we isolated CD14+ monocytes from the blood of healthy donors, cultured them in the presence of GM-CSF and M-CSF to differentiate them to M1 and M2 macrophages respectively.
Culture was performed under normoxia at 20% O2, hypoxia at 1% O2, and reperfusion – that is, growth at 1% with a 20% O2 feed.
At day 7 cells were stimulated with LPS overnight to respond.
On day 8 methanol-chloroform cell extraction was performed, with cell extracts analyses by 1D and 2D JRES NMR.
Analysis
NMR spectra (like this one shown here) were analysed using multivariate analysis techniques, including partial least squares discriminant analysis (PLS-DA) and partial least squares regression (PLS-R).
The former of these correlates elements of the spectra together, in this case peaks or metabolites, to describe the difference between two classification groups (e.g. M1 and M2). A sensitivity and specificity value for the resulting cross-validated model indicates the validity of separation between the groups.
The resulting loadings plot indicates the contribution of each peak, or metabolite, to this separation – either positively, or negatively.
PLS-R uses those same metabolites and peaks and attempts to find correlates of another continuous variable – such as a cytokine.
So, you take your list of metabolites, look at your metabolic pathway map, and figure out what it all means.
In case you remember anything from school, the TCA – or Kreb’s – cycle, is this one here. For the rest of it, you’re on your own.
This is quite hard.
Metapath
To simplify the process of deriving meaning from data – key to this sort of ‘systems biology’ – I developed software to enable evaluation of metabolic change within a pathway context.
I rather unimaginatively called it MetaPath.
The software is based on the MetaCyc collection of databases, and incorporates metabolites, genes, reactions for the analysis of metabolomic, transcriptomic and proteinomic data.
It uses a number of pathway analysis approaches, based on graph theory, to determine the importance of elements of a network – such as connectedness, centrality – together with completeness and overall change. Taking this it performs automated ‘pathway mining’ to identify the key locations of regulation.
The end result is a map that looks – something – like this. Red is up, blue is down, grey unchanged, white no-data. Mining is performed using PLS-DA weights, while each shade indicates a log2 increase in concentration from raw data.
So, we’ll now apply these methods to the data derived from our experiment.
M1 & M2 activation
Here we see the metabolic consequences of M1 and M2 macrophage activation by LPS under normoxia at 20%.
M1s show an ATP dependent response with upregulation of TCA cycle metabolism. The exception 2-oxoglutarate is a common metabolite to 25 reactions and pathways - indicated by the thick border – and is influence by factors outside the TCA.
In contrast M2 activation is heavily amino acid dependent with increases in amino acids across the board, indicative of turnover of essential amino acids, and synthesis, turnover of non-essential; 3-phospho-D-glycerate is a precursor to serine biosynthesis and depleted from the glycolytic pathway. Pyruvate is raised as a downstream byproduct of all multiple pathways, but levels exceed M2 TCA capability and it is excreted as Lactate.
M1 hypoxia
Now focusing solely on M1s, under hypoxia at 1% O2.
Here we see an expected down-regulation of the oxygen-dependent TCA cycle. In contrast creatinine-phosphate uptake is increased, a key anaerobic donator of phosphate to form ATP.
Alternate energy sources are being sought, with the reduction in essential amino acids suggesting consumption.
Leucine is ketogenic and can be turned into acetyl-coA to feed the TCA cycle producing ketone body acetoacetate in the process. Similarly, carnitine, essential for the transport of fatty acids to the mitochondria is also raised, but both this energy sources are oxygen dependent, so it’s not going to work.
Overall, what we see is a cell successfully utilizing alternative energy sources – as well as effectively stockpiling –currently- unusable fuel.
M1 reperfusion
Under reperfusion conditions – culture at 1% O2 with 20% O2 feed – we see a dramatic upregulation of the TCA, mimicking LPS stimulation – utilizing the limited, intermittent O2.
Creatine phosphate is raised as under hypoxia, as an alternative energy source.
Unfortunately, while I can write fancy software, I haven’t yet solved the problem of getting a picture to fit on a Powerpoint slide – these pathways extend off here & here – so I’ve summarized them in a table.
As you can see, aside from the central metabolism pathways, the majority of change is involved in amino acid breakdown – figures indicate sum log2 change (in this case overall entirely positive, with the exception of 2-oxoglutarate).
Functional correlates (PLS-R)
To relate these changes to something more familiar, we have begun to look for functional correlates of metabolic change – in this case IL-6 production. Here metabolic profiles have been correlated against IL-6 production by activated macrophages using PLS-R
The result is a model that can predict IL-6 from a given metabolic profile. High scoring metabolites are those that are consistently upregulated in IL-6 production, and may tell us something about the underlying metabolic mechanisms of cell activation.
As shown here the two key pathways correlating with IL-6 production are the TCA cycle, and the urea cycle - reflecting energy production, and amino acid breakdown respectively – confirming our earlier findings.
Conclusions
To conclude, we have shown
Anti- and pro-inflammatory macrophages have distinct metabolism under activation – with M1s ATP dependent, and M2s amino acid dependent.
Hypoxia drives catabolism, amino acid consumption & attempts to utilize alternate energy sources.
Reperfusion injury is a strong metabolic stimulant, equivalent to LPS.
That leads to the question – Can we use this knowledge of metabolic pathway regulation to tip the balance from pro- to anti- inflammatory macrophages in the rheumatoid synovium?
Acknowledgements
I would like to thank my supervisors Steve Young and Graham Wallace; the members of our research group and the Rheumatology Research Group as a whole; and also John Byrne and Mark Viant from University of Birmingham biosciences for the PLS-R analysis and 2D NMR metabolite identification.
Finally, I would also like to thank my funders the Wellcome Trust without whom this research would not have been possible.