Monday, 10 November 2014

Metabolomics and autism: the continuing search for biomarkers

I'm always a happy bunny when some of my own research findings receive something like independent replication. So it was when I read the monster paper from Paul West and colleagues [1] (open-access) reporting results based on not one, not two, not three, not even four, but five mass spectrometric methods looking for potential biomarkers for autism. Metabolomics in action (see here for an introduction to this topic).

Rosina @ Wikipedia 
The particular reason for my excitement was the quote: "Creatinine was decreased in children with ASD [autism spectrum disorder] and is consistent with the findings of Whitely et al, who observed similar changes in urinary creatinine in children diagnosed with PDD [pervasive developmental disorder]" based on the results of our paper a few years back [2] (see here for further details of how I spend my spare time talking about urine). The caveat being that we (the Royal We) looked in urine and West et al looked at blood (plasma). I can also forgive the authors for spelling my name wrong too... WHITELEY.

The West paper is open-access but I'm gonna give you a few pointers nonetheless. Stick with me on this one because although quite a long post, this is important work...

  • So: "The aim of the study was to perform a broad evaluation of small molecules in blood plasma to discover metabolites that may lead to biomarkers associated with ASD." 
  • The value-added bit was that this was study from the MIND Institute which meant that participant groups were very well-defined in terms of diagnosis and presenting symptoms. Indeed, as per other studies of biomarkers (see here), the talk was all about study groups (ASD vs. asymptomatic controls denoted as 'typically developing' TD) and also the use of training and test sets, where: " 82 patient samples (52 ASD and 30 TD samples) were split into two sets, (1) a training set of 61 samples (39 ASD and 22 TD) for identification of statistically significant features and classification modeling and (2) a 21-sample independent validation set (13 ASD and 8 TD) used to evaluate performance of the classification models."
  • So, then to the interesting bit... the mass spec methods used and data handling. A combination of liquid chromatography-high resolution mass spectrometry (LC-HRMS) and gas chromatography-mass spectrometry (GC-MS) were used. Actually the LC-HRMS was based on separation using C8 and HILIC column chromatography (the LC part) on both occasions coupled to "electrospray ionization" (the MS part) in positive and negative ion mode so giving "4 separate data acquisitions per sample." That and the GC-MS data makes 5 methods. 
  • Various methods/software were used to identify potential metabolites of interest including a couple of programs we use in our lab such as "Agilent Technologies MassHunter Qualitative Analysis software" and the METLIN database.
  • Results: as one might imagine, quite a few compounds/metabolites/signals were picked up across the 5 methods used. Table 2 of the paper gives you some idea of the sorts of numbers talked about. That being said, assigning a molecular formula to all those metabolites is rather another matter as per the authors note: "... 179 features comprised 3% of the LC-HRMS and 8% of the GC-MS preprocessed set of features". 'Features' by the way referred to "a moiety detected by the mass spectrometer that is defined by 2 properties 1) the detected mass-to-charge ratio (m/z) and 2) the chromatographic retention time".
  • Those 179 'features' formed the basis of the statistical analyses used to try and differentiate autism from control samples. These were subsequently whittled down to: "an 80 feature set [that] exhibited the best combined classification performance metrics... with an average accuracy of 90%, an average sensitivity of 92%, an average specificity of 87%, and an average AUC [area under the curve] of 0.95."
  • When moving from training to validation sets, the previous 80 feature model did not work as well. Indeed, some further statistical modelling was used and: "The results suggest that at least 40 features are needed to reach an accuracy of 70% and that a range of 80 to 160 features had the best performance with this independent validation sample set as well as the training set of samples."
  • To get to the juicy details of which compounds might be the ones to watch with autism biomarkers in mind, well: "a variety of molecular classes including amino acids, organic acids, sterols, and fatty acids" came up. I've already mentioned creatinine but other prominent mentions were given to "aspartate, glutamate, DHEAS, citric acid, succinic acid, methylhexa-, tetra- and hepta-decanoic acids, isoleucine, glutaric acid, 3-aminoisobutyric acid" and homocitrulline. The authors provide a handy overview of where their results might fit with other autism research areas (e.g. mitochondrial dysfunction, the gut microbiome) which I would encourage interested readers to further peruse. I'm gonna highlight isoleucine as one example where a form of autism has already been talked about with the words 'branched chain amino acids' in mind (see here).
  • The authors conclude with a need for quite a bit more study in this area: "This initial study provides proof of concept to further pursue development of metabolic biomarkers of ASD." Personally, I'd like to think that proof-of-concept is perhaps too preliminary a way of introducing metabolomics to autism research given previous research forays (see here and see here and see here) and their potentially important findings. Certainly, things need to be scaled up in terms of participant numbers [3] and also delving into those all-important subgroups of 'the autisms'. Challenges however do remain in assigning molecular formulae to all those compounds detected.

What's more to say? Well, as has been mentioned in a previous post (see here) one always needs to be a little careful when talking about biomarkers for autism as if we're talking about a homogeneous diagnosis and the search for compound X supposedly covering all that heterogeneity (and comorbidity). If we've learned anything from the genetic research on autism for example, it is that simple, universal objective markers are probably not going to be present. Given that the metabolome is to quite a large extent determined by the proteome potentially also intersecting with the microbiome, complexity is probably going to be the keyword.

That being said, I do see merit in the continued use of metabolomics as part of all that system biology kerfuffle (see here) when applied to autism research. I'd personally suggest a few tweaks to how this kind of research is carried out on the basis for example, of not necessarily using the diagnostic label of 'autism' or 'autism spectrum disorder' as a primary starting point. I've talked before on this blog about the notion of best responders and non-responders to intervention for example (see here) and how if one chose to use this as an important variable differentiating those on the spectrum, one might just see a few differences across groups. Such research might also help further inform researchers / clinicians / parents / people on the spectrum who might be best suited for certain types of intervention. Interestingly with dietary intervention in mind, the authors reported that: "Ten of the 52 ASD subjects were on a gluten and/or casein-free (GFCF) diet". Mmm...

Then to some music... Go Your Own Way by Fleetwood Mac. Or if you prefer, the Seaweed version...


[1] West PR. et al. Metabolomics as a Tool for Discovery of Biomarkers of Autism Spectrum Disorder in the Blood Plasma of Children. PLoS One. 2014 Nov 7;9(11):e112445.

[2] Whiteley P. et al. Spot urinary creatinine excretion in pervasive developmental disorders. Pediatr Int. 2006 Jun;48(3):292-7.

[3] Roessner V. Large sample size in child and adolescent psychiatric research: the way of salvation? European Child & Adolescent Psychiatry. 2014. November 6.

---------- West, P., Amaral, D., Bais, P., Smith, A., Egnash, L., Ross, M., Palmer, J., Fontaine, B., Conard, K., Corbett, B., Cezar, G., Donley, E., & Burrier, R. (2014). Metabolomics as a Tool for Discovery of Biomarkers of Autism Spectrum Disorder in the Blood Plasma of Children PLoS ONE, 9 (11) DOI: 10.1371/journal.pone.0112445