Detection of malignant tumours by multivariate analysis of proton magnetic resonance spectra of serum

MRS-based Metabolomics in Cancer Research

Metabolomics is a relatively new technique that is gaining importance very rapidly. MRS-based metabolomics, in particular, is becoming a useful tool in the study of body fluids, tissue biopsies and whole organisms. Advances in analytical techniques and data analysis methods have opened a new opportunity for such technology to contribute in the field of diagnostics. In the MRS approach to the diagnosis of disease, it is important that the analysis utilizes all the essential information in the spectra, is robust, and is non-subjective. Although some of the data analytic methods widely used in chemical and biological sciences are sketched, a more extensive discussion is given of a 5-stage Statistical Classification Strategy. This proposes powerful feature selection methods, based on, for example, genetic algorithms and novel projection techniques. The applications of MRS-based metabolomics in breast cancer, prostate cancer, colorectal cancer, pancreatic cancer, hepatobiliary cancers, gastric cancer, and brain cancer have been reviewed. While the majority of these applications relate to body fluids and tissue biopsies, some in vivo applications have also been included. It should be emphasized that the number of subjects studied must be sufficiently large to ensure a robust diagnostic classification. Before MRS-based metabolomics can become a widely used clinical tool, however, certain challenges need to be overcome. These include manufacturing user-friendly commercial instruments with all the essential features, and educating physicians and medical technologists in the acquisition, analysis, and interpretation of metabolomics data.

Pretreatment of Mass Spectral Profiles: Application to Proteomic Data

Mass spectral profiles are influenced by several factors that have no relation to compositional differences between samples: baseline effects, shifts in mass-to-charge ratio (m/z) (synchronization/alignment problem), structured noise (heteroscedasticity), and, differences in signal intensities (normalization problem). Different procedures for pretreatment of whole mass spectral profiles described by almost 50,000 m/z values are investigated in order to find optimal approaches with respect to revealing the information content in the data. In order to quantitatively assess the impact of different procedures for pretreatment of mass spectral profiles, we use factorial designs with the ratio between intergroup and intragroup (replicate) variance as response. We have examined the influence of smoothing, binning, alignment/synchronization, noise pattern, and normalization on data interpretation. Our analysis shows that the spectral profiles have to be corrected for heteroscedastic noise prior to normalization. An nth root transform, where n is a small, positive integer, is used to create a homoscedastic noise structure without destroying the linear correlation structures describing individual components when using whole mass spectral profiles. The choice of n is decided by a simple graphic procedure using replicate information. Log transform is shown to change the heteroscedastic noise structure from being dominant in high-intensity regions, to produce the largest noise in the low-intensity regions. In addition, log transform has a negative effect on the collinearity in the profiles. Factorial designs reveal strong interactions between several of the pretreatment steps, e.g., noise structure and normalization. This underlines the limited usability of looking at the different pretreatment steps in isolation. Binning turns out to be able to substitute smoothing of spectra by, for example, moving average or Savitsky-Golay, while, at the same time, reducing the data point description of the profiles by 1 order of magnitude. Thus, if the sampling density is high, binning seems to be an attractive option for data reduction without the risk of losing information accompanying the integration of profiles into peaks. In the absence of smoothing, binning should be executed prior to alignment. If binning is not performed, the order of pretreatment should be smoothing, alignment, nth root transform, and normalization.

Quantification of plasma lipids and apolipoproteins by use of proton NMR spectroscopy, multivariate and neural network analysis

New approaches for quantification of human blood plasma lipids and apolipoproteins are presented. One method is based on multivariate analysis of proton nuclear magnetic resonance spectra of human blood plasma. Although similar approaches have been developed previously, this is the first time principal component analysis (PCA) and partial least squares regression (PLS) have been applied to this particular task. Further, a large proportion of the subjects in this study were cancer patients undergoing treatment, which introduced a new dimension to the quantification of lipoprotein distributions. Calibration models for prediction of lipids and apolipoproteins were constructed by use of PLS, and blind samples were used to test the predictive ability. Comparison of the predicted vs observed data obtained by standard clinical chemical procedures gave good agreement; the correlation coefficient for total plasma triglyceride was 0.99, for total plasma cholesterol 0.98, for LDL cholesterol 0. 97, and for HDL cholesterol 0.88. These results are comparable with those obtained with other methods. The quantitative analysis of 14 components (including total cholesterol and total triglyceride) of human blood plasma was also undertaken using various neural network (NN) analyses of selected portions of the spectra. Conventional fully connected backpropagation neural network topologies were capable of providing excellent predictions for the majority of the variables, confirming and reinforcing literature related to this approach. However HDL triglycerides were poorly predicted, while intermediate-quality results were obtained for the LDL cholesterol, plasma apoA1 and LDL apoB variables. In these instances, applying significantly different neural network algorithms involving either general regression or polynomial neural networks in combination with genetic adaptive components for parameter optimisation made improved predictions.

Classification of CSF samples from normal and post-partum psychotic women using chromatographic profiles with bilinear projections: a multivariate approach

This paper demonstrates how chromatographic profiles of cerebrospinal fluid (CSF) have been subjected to multivariate data analysis to discriminate between CSF samples from women with post-partum psychosis and those from healthy women. Instead of peak-heights or areas, digitally defined chromatographic profiles were examined using principal component analysis (PCA). In accordance with the diagnosis, we have found a complex profile pattern of at least ten composite peaks that discriminates between these groups. Two of these peaks were for the discrimination particularly clearly between the two groups. We speculate that these findings can be useful in the diagnosis of post-partum psychosis, increasing diagnostic precision and having both clinical and prognostic implications.

Proton nuclear magnetic resonance spectroscopy of serum from patients with colorectal neoplasia

Proton nuclear magnetic resonance (NMR) spectra of serum have been recorded from patients with colorectal neoplastic polyps, before and after treatment of colorectal cancer, in patients with advanced lung cancer, and also from healthy controls. Digitally defined NMR profiles of the methyl and methylene peaks were used as input for supervised principal component modelling. An unknown sample was classified according to its residual, i.e. the difference between the spectral pattern of the unknown and control group. There was a statistically significant difference between the mean residual in the untreated colorectal cancer group and in controls (P = 0.003). The sensitivity of detecting untreated colorectal cancer was only 20%. There were no stage-dependent differences between the residuals within the untreated colorectal cancer group. After curative surgery, four patients had recurrence of malignant disease without an increase in residual prior to recurrence. Patients with advanced malignant disease (lung cancer WHO stage IIIB and IV) had a highly significant difference in mean residual from that of controls, with a sensitivity of detecting cancer of 87.5%. This increase in residual could not be explained by increase in the level of serum triglyceride. NMR spectroscopy was not a useful diagnostic tool in patients with colorectal neoplastic polyps and cancer.

Nuclear magnetic resonance spectroscopy of plasma to distinguish between malignant and benign diseases causing jaundice and cholestasis

The sera of 51 patients with malignant (n = 25) and benign (n = 26) hepatopancreatobiliary disorders were analysed by 1H magnetic resonance spectroscopy (NMR) in order to distinguish between malignant and benign diseases causing jaundice and/or cholestasis. Macromolecular linewidths were determined both manually and automatically with a computed analysis, and both methylene (CH2) and methyl (CH3) resonances were evaluated. The mean linewidth of the CH3 peak was significantly narrower in the patients with malignant disease than in the patients with benign disease both in the manual and computed analyses, but no significant differences in the CH2 peak were detected. Diagnostic sensitivity and specificity of the CH3 peak determined in the computed analysis were 92% and 27% respectively. In the light of the current study, it seems obvious that because overlap between benign and malignant groups was too great, 1H NMR spectroscopy of plasma is not of practical value in distinguishing between benign and malignant causes of jaundice and/or cholestasis.

In vitro characterization of lung cancers by the use of 1H nuclear magnetic resonance spectroscopy of tissue extracts and discriminant factor analysis

Using proton magnetic resonance spectroscopy (1H MRS) spectra were obtained in vitro from extracts of four types of lung cancer (squamous cell, adenocarcinoma, large cell, small cell) and normal lung. The hydrophilic phase of the chloroform/methanol-water extracts yielded several distinct peaks. Among them the peak areas for cholines, creatines, glycine, and alanine, and their ratios were calculated and used as parameters to characterize different lung tissues. The ratios, cholines/alanine and glycine/alanine, were significantly (P < 0.001 to P < 0.05) higher for the normal lung than lung cancers. Creatines/glycine and creatines/cholines generally provided good discrimination (P < 0.001 to P < 0.05) between any two types of lung cancer. When data were further analyzed by discriminant factor analysis, there was 81.5 to 90.7% accuracy in predicting between normal lung and each cancer type, or among the four types of lung cancer. These results suggested that 1H MRS might be useful as an adjunct modality in the differential diagnosis of lung cancers.

C-13 NMR spectroscopy of plasma reduces interference of hypertriglyceridemia in the H-1 NMR detection of malignancy. Application in patients with breast lesions

We have previously described the application of water-suppressed proton nuclear magnetic resonance (H-1 NMR) spectroscopy of plasma for detection of malignancy. Subsequently, hypertriglyceridemia has been identified as a source of false positive results. We now describe a confirmatory, adjunctive technique--analysis of the carbon-13 (C-13) NMR spectrum of plasma--which also identifies the presence of malignancy but is not sensitive to the plasma triglyceride level. Blinded plasma samples from 480 normal donors and 208 patients scheduled for breast biopsy were analyzed by water-suppressed H-1 and C-13 NMR spectroscopy. Triglyceride levels were also measured. Among the normal donors, there were 38 individuals with hypertriglyceridemia of whom 18 had results consistent with malignancy by H-1 NMR spectroscopy. However, the C-13 technique reduced the apparent H-1 false positive rate from 7.0% to 0.6%. Similarly, in the breast biopsy cohort, C-13 reduced the false positive rate from 2.8% to 0.9%. Furthermore, the accuracy of the combined H-1/C-13 test in this blinded study was greater than 96% in 208 patients studied.