Final Activity Report Summary - AUTMETABOL (Chemical markers for autism: a metabolite profiling study)
The main scientific achievements of the project have been: (i) to develop novel methods for pre-processing of data obtained using micellar electrokinetic capillary chromatography (MEKC) to analyse urine samples; (ii) to apply these methods to metabolic fingerprinting of urine samples from autistic and control groups.
The pre-processing methods are generic, should be applicable to other analytical techniques and other urine samples, and are currently being extended to results obtained by proton nuclear magnetic resonance (1H-NMR) and liquid chromatography - mass spectrometry (LC-MS) on the same sample set.
Autism is a condition involving abnormal childhood neurological developmental behaviour. Whilst there are accepted diagnostic criteria based on behavioural assessment, there is at present no consensus on biochemical factors involved in this chronic condition and no universally-accepted diagnostic markers for autism have been found yet. The application of metabolic fingerprinting in a non-targeted approach allows an exploratory rather than hypothesis-driven investigation into the differences between groups and does not restrict the analysis to a single compound or group of compounds. The application of non-targeted approaches presents difficulties associated with the data collection from a large number of compounds present in different concentrations. However, the limiting stage in metabolic fingerprinting is the data analysis. Multivariate methods for data analysis are needed with the large data sets required to reach conclusions of statistical validity. Various pre-processing techniques must be applied in order to obtain suitable variables for pattern recognition and classification. These pre-processing techniques eliminate the effects of factors such as noise, baseline drift, differences in concentration and peak misalignment. In this study, urine samples from 49 children on the autism spectrum and 125 healthy children (control group) were analysed in a non-targeted and unbiased search for markers for autism. The analysis method involved separation and detection of components by MEKC-UV.
The novel combination of data pre-processing methods for the MEKC-UV electropherograms prior to their statistical analysis included the following: (i)Denoising: The use of non-decimated wavelet transform with the Haar filter showed excellent results in the removal of jagged features while tracking the peak profiles in the urinary electropherograms. (ii) Normalisation: Creatinine concentrations determined by MEKC-UV were used to compensate for variations in urine concentration prior to further processing. (iii) Baseline correction: Urinary electropherograms contain many peaks and few areas can be unambiguously identified as containing only noise. This makes fitting a curve to the baseline or using wavelet approximations inadvisable. Fitting the baseline as linear segments was chosen instead. (iv) Alignment: A major obstacle for the comparison of CE generated metabolic profiles is migration time variability. We combined alignment to reference peaks with a novel pair-wise alignment strategy involving spectral information from multiple wavelengths.
Although no bio-markers for autism could be determined from the application of PCA-LDA and PLS-LDA to the MEKC-UV urinary profiles (probably because of the natural variability of the groups of samples compared), the data pre-processing algorithms developed in this project show the advantage that they can be applied to the processing of other data collected by CE-UV.
As the relationship of serotonin to autism has been investigated extensively, but the implications remain unclear due to conflicting findings, an alternative directed strategy based on the analysis of serotonin metabolites by HPLC-MS was carried out in parallel to the research described above. Results on this last phase of the project are currently being processed.
The pre-processing methods are generic, should be applicable to other analytical techniques and other urine samples, and are currently being extended to results obtained by proton nuclear magnetic resonance (1H-NMR) and liquid chromatography - mass spectrometry (LC-MS) on the same sample set.
Autism is a condition involving abnormal childhood neurological developmental behaviour. Whilst there are accepted diagnostic criteria based on behavioural assessment, there is at present no consensus on biochemical factors involved in this chronic condition and no universally-accepted diagnostic markers for autism have been found yet. The application of metabolic fingerprinting in a non-targeted approach allows an exploratory rather than hypothesis-driven investigation into the differences between groups and does not restrict the analysis to a single compound or group of compounds. The application of non-targeted approaches presents difficulties associated with the data collection from a large number of compounds present in different concentrations. However, the limiting stage in metabolic fingerprinting is the data analysis. Multivariate methods for data analysis are needed with the large data sets required to reach conclusions of statistical validity. Various pre-processing techniques must be applied in order to obtain suitable variables for pattern recognition and classification. These pre-processing techniques eliminate the effects of factors such as noise, baseline drift, differences in concentration and peak misalignment. In this study, urine samples from 49 children on the autism spectrum and 125 healthy children (control group) were analysed in a non-targeted and unbiased search for markers for autism. The analysis method involved separation and detection of components by MEKC-UV.
The novel combination of data pre-processing methods for the MEKC-UV electropherograms prior to their statistical analysis included the following: (i)Denoising: The use of non-decimated wavelet transform with the Haar filter showed excellent results in the removal of jagged features while tracking the peak profiles in the urinary electropherograms. (ii) Normalisation: Creatinine concentrations determined by MEKC-UV were used to compensate for variations in urine concentration prior to further processing. (iii) Baseline correction: Urinary electropherograms contain many peaks and few areas can be unambiguously identified as containing only noise. This makes fitting a curve to the baseline or using wavelet approximations inadvisable. Fitting the baseline as linear segments was chosen instead. (iv) Alignment: A major obstacle for the comparison of CE generated metabolic profiles is migration time variability. We combined alignment to reference peaks with a novel pair-wise alignment strategy involving spectral information from multiple wavelengths.
Although no bio-markers for autism could be determined from the application of PCA-LDA and PLS-LDA to the MEKC-UV urinary profiles (probably because of the natural variability of the groups of samples compared), the data pre-processing algorithms developed in this project show the advantage that they can be applied to the processing of other data collected by CE-UV.
As the relationship of serotonin to autism has been investigated extensively, but the implications remain unclear due to conflicting findings, an alternative directed strategy based on the analysis of serotonin metabolites by HPLC-MS was carried out in parallel to the research described above. Results on this last phase of the project are currently being processed.