Characterization of the serum metabolome following radiation treatment in patients with high-grade gliomas

Eleven patients with radiologic suspicion of a high grade glioma not accessible for surgical resection were included in this study. All patients underwent a stereotactic biopsy to obtain a histopathological diagnosis. After confirmation (10 GBM and one astrocytoma III), microdialysis catheters were implanted with stereotactic technique, one into the contrast-enhancing tumor tissue and one approximately 10 mm outside of the tumor in brain-adjacent to tumor (BAT) [14]. One catheter was put in the abdominal subcutaneous tissue as reference. All catheters had a 10 mm long semi-permeable membrane with a cutoff of 100 kDa (CMA 71, CMA Microdialysis, Stockholm, Sweden). The catheters were perfused with Ringer solution (Perfusion fluid T1; CMA Microdialysis) mixed with Dextran (30 g Dextran 60 1000 mL-1), to prevent microfiltration, and with a flow-rate of 0.3 μl/min (CMA 106 or 107, CMA Microdialysis).

All patients were treated with radiotherapy started within 2 to 5 days after surgery. A standard schedule of 2 Gy × 30 were given to eight of the patients. Three patients in poor general condition were treated using hypofractionationated schedules 3 Gy × 13 for two patients, and 3.4 Gy × 10 for one patient. Fasting serum samples were collected at admittance to the ward, on the same day but before the first treatment session and in the morning after the second and fifth radiotherapy fraction (Table 1). The extracellular compartment were collected every second hour from surgery to the morning after the fifth radiotherapy fraction. Three samples for analysis were selected before the first radiation dose and samples for analysis were matched to the same time-span as for the serum samples. The results from the extracellular compartment in tumor tissue and BAT in the same group of patients have been reported earlier [13].

All patients participated voluntarily and gave their fully informed consent. The study was approved by the Ethics committee of Umeå University.

The serum samples were thawed in room temperature for 30 min before addition 900 μl extraction solution consisting of methanol (90 %) and water (10 %) with 11 internal standards (IS) (7 ng/μl) to 100 μl serum. The samples were extracted using a MM301 vibration Mill (Retsch GmbH & Co. KG, Haan, Germany) for 2 min, 30 Hz, after 2 h on ice, centrifuged 15 min, 4 °C, 14,000 rpm. Then 200 μl of the collected supernatant was transferred to vials and evaporated to complete dryness before methoxymation with 30 μl of methoxyamine solution in pyridine (15 μg/μl) first at 70 °C for 1 h then in room temperature for 16 h. Thereafter, the samples were trimethylsilylated with 30 μl of MSTFA at room temperature for 1 h and before addition of 30 μl of heptane (containing 0.5 μg of methyl stearate).

GC-TOFMS was used to screen the serum samples for a broad spectrum of metabolites in terms of chemical properties. In this way a robust common metabolite profile can be obtained for all samples and used for further multivariate sample comparisons. Prior to analysis, the samples were randomized and analyzed together with a series of n-alkanes (C₁₂-C₃₂) to allow retention indexes to be calculated. 1 μl sample was injected splitless by an Agilent 7683 Series autosampler (Agilent, Atlanta, GA) into an Agilent 6980 GC equipped with a 10 m × 0.18 mm i.d. fused-silica capillary column chemically bonded with 0.18 μm DB5-MS stationary phase (J&W Scientific, Folsom, CA). The injector temperature was set at 270 °C. Helium was used as carrier gas at a constant flow rate of 1 ml/min through the column. The purge time was set to 60 s at a purge flow rate of 20 ml/min and an equilibration time of 1 min for every analysis. Initially, the column temperature was kept at 70 °C for 2 min and then increased to 320 °C at 30 °C/min, where it was kept for 2 min. The column effluent was introduced into the ion source of a Pegasus III TOFMS (Leco Corp., St Joseph, MI). The transfer line temperature was set at 250 °C and the ion source temperature at 200 °C. Ions were generated by a 70 eV electron beam at a current of 2.0 mA. Masses were acquired from m/z 50 to 800 at a rate of 30 spectra/s, and the acceleration voltage was turned on after a solvent delay of 165 s. The acquired data was exported to MATLAB 7.11.0 (R2010b) (Mathworks, Natick, MA) as NetCDF files. MATLAB 7.3 (R2006b) (Mathworks, Natick, MA) was used for data processing for the MD samples.

Metabolomics screening by means of GC-TOFMS generates a profile of more or less overlapping compounds (metabolites). To be able to obtain a reliable quantification and identification of the detected metabolites for further sample analyses and evaluation the raw GC-TOFMS data has to be resolved into pure chromatographic peaks and mass spectra for each individual metabolite. To achieve this, baseline correction, alignment, time-window settings and hierarchical multivariate curve resolution (HMCR) [15, 16] were performed using in-house developed scripts in Matlab 7.11.0 (R2010b) (Mathworks, Natick, MA). The chromatograms were divided into 56 time windows from which the chromatographic peaks were resolved resulting in a data matrix (X) were each row represents one serum sample and each column represents one metabolite. For each metabolite in each sample the area under the chromatographic peak, the relative concentration, is calculated. All peak areas were normalized using the peak areas from the 11 internal standards. To identify the detected compounds the mass spectral profile and retentions index were compared with spectra in an in-house spectra library, the NIST library 2.0 (as of January 31, 2001), and the mass spectra library maintained by the Max Planck Institute in Golm. This was followed by a manual inspection and curation of the data to further resolve co-eluting compounds and to correct for split peaks.

The pre-treated and HMCR resolved GC-TOFMS data describes the complex interactions taking place in the serum metabolome of GBM patients as a consequence of radiotherapy. To be able to extract information of patterns of co-varying metabolites associated with the effect of radiotherapy methods that can handle variable correlations and allows interpretation of such complex interactions are required. To achieve this, chemometric bioinformatics by means of multivariate projection methods was applied to the processed data in order to detect and evaluate metabolite patterns in serum associated with the effect of radiotherapy. Initially, principal component analysis (PCA) [17] was applied to the data to obtain an overview of the variation in the data and detect possible outliers. Outlier detection was performed in both the PCA model space (scores) as well as in the model residual (distance to model in X (DModX)) and to qualify as an outlier leading to exclusion from the analysis a sample show a clearly deviating pattern in either of these two measures. Furthermore, orthogonal partial least squares-discriminant analysis (OPLS-DA) [18] was used to establish if there was a systematic difference in serum metabolite profile between samples collected before and after treatment.

To be able to further analyze treatment induced changes on an individual patient level, the data was pre-treated with the same approach as in our previous study on the same patients, using the Individual Treatment Over Time (ITOT) normalization [13]. The first two samples from the same patient were categorized as untreated and the last two samples were categorized as treated. The value from the first time point (4 days before treatment) was subtracted from the first two time points (including itself). The original value from time point 2 was then included as the first value in the treated samples and subtracted from itself as well as from time point 3 and 4 resulting in that both the treated and the untreated group were set to have the same starting point, for each patient. Patient one, missing its two last time points, was excluded from the analysis. Patient 5 was also excluded because time point two was missing making it impossible to normalize its treated samples.

The ITOT normalized data was then modeled by OPLS-DA with treatment as the response variable (y). OPLS-DA is a supervised projection method that describes the relations between the descriptive data (X) and the response variable (y) by dividing the systematic variation in X in two parts, one related to y, predictive and the other unrelated, orthogonal, to y [18]. By separating the systematic variation into predictive and orthogonal components the interpretation of the model is greatly facilitated. Variables with low model weight values (|w*| < 0.05), variables unaffected by treatment, were discarded from the modelling.

A p-value based on ANOVA on the cross-validated OPLS-DA models was used for model validation. The individual metabolites were tested for significance using a Student’s t-test, where p-values < 0.05 were considered to be significant. To be able to investigate the whole metabolite pattern in serum compared to the whole metabolite pattern in the extracellular compartment in tumor from our previous study, the cross-validated score values for each model were compared in the same analysis.

All patients did undergo the stereotactic procedure and the following period of microdialysis and radiotherapy without any procedure-related side effects or complications.

From the acquired serum GC-TOFMS data 159 putative metabolite peaks were resolved using HMCR. Of these 58 (36 %) could be identified by means of library comparisons. The unidentified metabolites still retained information regarding retention time index and fragmentation patterns and were thus kept in the analysis. This can be compared to the extracellular compartment from tumor tissue where, 151 metabolites were reliably detected and quantified and 67 out of those metabolites were identified (44 %) [13].

No outliers were found in the initial PCA analysis of the HMCR processed data (not shown). As a first modelling step to verify possible treatment specific alterations of the serum metabolite profile OPLS-DA was applied directly on the HMCR processed data. The cross-validated score plot from the calculated OPLS-DA model (Fig. 1) show that there is a general systematic difference in metabolite profile between serum samples taken before and after treatment (p = 0.01). To further investigate the patient specific response to treatment the ITOT normalized data was analyzed using OPLS-DA with treatment as the response variable. Based on the model weights, variables were selected and a new OPLS-DA model was calculated. The cross-validated score plot from the final OPLS-DA model (Fig. 2) indicate that the majority of the patients show a metabolic response to radiotherapy in serum (p = 0.006). Notably, patients 6 and 11 show a more moderate or no response to the treatment compared to the other patients, indicating a possibility to detect and follow individual differences of treatment in serum close to real-time.

In serum, 84 metabolites passed the cut off criteria (|w*| > 0.05) (Table 2) and 25 of them also showed a significant univariate p-value (<0.05). Sixteen metabolites showed higher levels during and after treatment as compared to before treatment while 68 metabolites showed lower levels (Table 2). Out of the identified metabolites, creatinine, threonine, glyceric acid, tyrosine, oleic acid and methionine had the strongest correlation to treatment with lower serum levels after treatment.

The metabolomic findings in extracellular fluid from tumor tissue and BAT from our previous report, in the very same patients, have been included in Additional file 1: Table S1 for comparison. When comparing the results we found metabolites that were affected by treatment in both serum and extracellular fluid from both tumor tissue and BAT. Ornithine, tyrosine and urea decreased in both serum and tumor. Glutamine and glutamate decreased in serum but increased in tumor tissue and BAT. Citric acid increased in serum and BAT while creatinine and glyceric acid decreased in serum samples but showed an increase in BAT.

When interpreting the cross-validated score plots from the modeled data, built on all metabolites that passed the cut off criteria, for both the serum and extracellular compartment from the tumor (Fig. 3) we can see that the pattern before and during treatment is the same or similar for all patients irrespective of compartment. This indicates that the systematic response to treatment in the tumor is reflected in serum even if not all the individual metabolite levels coincide.