¹H NMR-based metabolic profiling of human rectal cancer tissue

Typical ¹H NMR spectra of aqueous phase extracts of rectal cancer tissues of different stages and normal mucosae were shown in Figure 1. The standard one-dimension spectrum gave an overview of all metabolites. The major metabolites in the spectra were identified according to literature data and the Human Metabolome Database (http://​www.​hmdb.​ca/​). As a result, a series of changes in endogenous metabolite levels were observed in rectal cancer when compared with the normal mucosa. These metabolites included lactate, threonine, acetate, glutathione, uracil, succinate, serine, formate, lysine, tyrosine, myo-inositol, taurine, phosphocreatine, creatine, betaine, dimethylglycine, which are known to be involved in multiple metabolic processes, especially in energy and amino acid metabolism [14, 15].

To determine the differences between the normal controls and rectal cancer tissues, we initially utilized the PCA to analyze ¹H NMR data after data normalization. The results showed an apparent separation between rectal cancer tissues and normal controls on the scores plot of first two principal components (PC) (Figure 2A). The majority of samples were located in 95% confidence interval. Therefore, all of samples were used in the following analysis to ensure the maximum information.

To optimize the separation between the rectal cancer tissues and normal controls, of the two groups, we then utilized the OPLS-DA to visualize the metabolic difference. As shown in Figure 2B, good separation in the scores plot of PC1 and PC2 of OPLS-DA analysis was obtained between rectal cancer tissues and normal controls. Moreover, model parameters in the permutation test for the explained variation (R² = 0.89) and the predictive capability (Q² = 0.83) were significantly high, indicating a satisfactory predictive ability (Figure 2D). To identify the main metabolites responsible for the separation between cancer tissues and normal controls, their scores and loadings plots with correlation coefficients were obtained from OPLS-DA analysis based on the NMR data of tissue samples (Figure 2C). The loadings were colored according to the UV model variable weights and showed the significant class-discriminating metabolites responsible for the clustering patterns. The positive signals indicated the up-regulated metabolites in the cancer tissues in comparison with normal controls, including lactate, threonine, acetate, glutathione, uracil, succinate, serine, formate, lysine and tyrosine. On the other hand, the signals in the negative direction indicated the down-regulated metabolites in rectal cancer tissues, including myo-inositol, taurine, phosphocreatine, creatine, betaine and dimethylglycine. The significantly distinguishing metabolites were summarized according to VIP > 1 and p < 0.05 (Table 1). According to metabolic pathway on the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (http://​www.​genome.​jp/​kegg/​), we outlined the main metabolic pathways, which are closely related to rectal cancer morbidity. These metabolic pathways consisted of glycolysis, serine synthesis pathway, TCA cycle, amino acid metabolism, pyrimidine metabolism and gut flora metabolism.

The differences of metabolic profiling among various stages of rectal cancer are important for biomarker identification and for accurate molecular diagnosis and therapy. OPLS-DA analysis was applied to distinguish the metabolites difference between normal controls and each stage of rectal cancer tissues. The scores plots of PC1 and PC2 showed that all stages (I, II, III and IV) of rectal cancer tissues could be clearly distinguished from normal controls (Figure 3A). A panel of 40 metabolites with VIP > 1 from the training set and p < 0.05 from Student’s t-test were identified and summarized (Table 2). As shown in Table 2, creatine, uracil, succinate and β-hydroxybutyrate changed along with the process of the rectal cancer. Valine, lactate, glutamine, alanine, trimethylamine-n-oxide (TMAO), lysine and PC (phosphocholine) were increased in all rectal cancer patients except stage I. Interestingly, acetate, o-acetyl glycoprotein, dimethylamine and leucine became significantly different form stage III to stage IV. Moreover, NAD, formic acid, acetone, isoleucine, acetoacetic acid, sarcosine and acetoacetate were significantly up-regulated only in stage IV in comparison with normal controls. The corresponding loading plots based on OPLS-DA models were presented in Figure 3B. The color scale corresponded to the UV model variable weights. The relative changes in metabolites with significant correlation coefficients were a major discriminating factor among different populations, implying the biochemical alterations in different morbidity.

Model parameters of permutation analysis for different stages were as follows: stage I: R² = 0.93, Q² = 0.87; stage II: R² = 0.93, Q² = 0.86; stage III: R² = 0.93, Q² = 0.83 and stage IV: R2 = 0.95, Q2 = 0.83. These parameters indicated the excellence of the model (Figure 3C). To further confirm the performance of these models, 80% of samples were randomly selected as training samples. Prediction parameters of the remaining 20% of samples using OPLS-DA model established with the training samples: Normal vs stage I: R²X_cum = 0.179, R²Y_cum = 0.928, Q²Y_cum = 0.828; Normal vs stage II: R²X_cum = 0.19, R²Y_cum = 0.92, Q²Y_cum = 0.829; Normal vs stage III: R²X_cum = 0.16, R²Y_cum = 0.946, Q²Y_cum = 0.81; Normal vs stage IV: R²X_cum = 0.16, R²Y_cum = 0.941, Q²Y_cum = 0.754) (Figure 3D).

Biomarker identification is important for detecting the rectal cancer formation, invasion, and metastasis. The representative metabolites with significant difference between controls and rectal cancer tissues were represented in box-and-whisker plots (Figure 4), which showed the concentration ranges, median quartiles and extremes.

The decrease of glucose and increase of lactate in tumor tissues was not surprising because of the Warburg effect. The results of changes in myo-inositol and glucose in our work were consistent with a previous research on breast cancer [16]. Myo-inositol, a precursor in the phosphatidylinositol cycle and a source of several second messengers, was decreased along with the progression of rectal cancers compared with normal controls. The function of myo-inositol as an osmoregulator in different stages of malignant transformation could be a further explanation for our results.

Many studies have found free amino acids altered in patients with different kinds of cancer [14, 17]. In our study, leucine, glutamine, threonine and serine were significantly increased along with the progression of rectal cancer, which can be explained as cellular needs for higher turnover of structural proteins in cell proliferation. Sarcosine, an N-methyl derivative of the amino acid glycine, was significantly up-regulated in stage IV. Uracil is an indicator of transcription, whose increase suggests cell proliferation up-speeded. Methylamine, DMA and TMAO, the products of choline metabolism, were also altered in our study,indicating the disturbance of choline metabolism.

Based on the modified metabolites, we summarized a related metabolic pathway of rectal cancer. As shown in Figure 5, the disturbed metabolic pathway included glycolysis (glucose, lactate), tricarboxylic acid cycle (succinate), choline metabolism (TMAO, DMA, methylamine, betaine, dimethylgcine, sarcosine, creatine and phosphocreatine), ketone body (acetoacetate, β-hydroxybutyrate and acetone) and amino acid metabolism (serine and glycine).

A total of 127 rectal cancer patients were recruited from West China Hospital of Sichuan University during 2009 to 2010. The patients enrolled in this study did not receive any neoadjuvant chemotherapy or radiation therapy before surgical treatment. The clinical information of patients was summarized in Table 3. The rectal cancer tissue histology, tumor grade, TNM, Duke Stages was presented in Table 3. And the nutritional status of the patients, body weight and weight loss were provided (Additional file 1: Table S1). The survival rate of patients enrolled in this study was also provided (Additional file 2: Figure S1). As shown in Table 3, the stage of all tissue specimens was determined according with the American Joint Committee on Cancer (AJCC) for rectal tumors: stage I, 35 patients; stage II, 37 patients; stage III, 37 patients; stage IV, 18 patients. The protocols outlined in the following text were approved by the Ethics Committee of West China Hospital of Sichuan University. The informed consents were obtained from all patients prior to sample collection.

Tumor specimens and adjacent normal-appearing tissues at least 5–10 cm away from the edges of a tumor were collected from rectal cancer patients undergoing colorectal resection according to procedure reported previously [35]. In total, 170 tissue samples were obtained from patients. The tissues dissected by a senior pathologist in the operating room were immediately frozen in liquid nitrogen and stored at -80°C. The clinical diagnosis, tumor stage, histology differentiation and resection margin were determined by routine histopathology examination of H & E stained specimens by a blinded pathologist.

The frozen tissue samples ranged from 150 to 400 mg were weighed and suspended in methanol (4 ml per gram of tissue) and double distilled water (0.85 ml per gram of tissue). After vortex, chloroform (2 ml per gram of tissue) was added, followed by addition of 50% chloroform (2 ml per gram of tissue). The suspension was left on ice for 30 min, and centrifugated at 1,000 g for 30 min at 4°C. This procedure separated suspension into three phases: a water phase at the top, a denatured proteins phase in the middle, and a lipid phase at the bottom. The upper phase (aqueous phase) of each sample was collected and evaporated to dryness under a stream of nitrogen. The residue was reconstituted with 580 μl of D₂O containing 30 μM phosphate buffer solution (PBS, pH = 7.4) and 0.01 mg/ml sodium (3-trimethylsilyl)-2, 2, 3, 3-tetradeuteriopropionate (TSP) as a chemical shift reference (δ0.0). After centrifuged at 12,000 g for 5 min, the 550 μl supernatant was transferred into a 5-mm NMR tube for NMR spectroscopy [36].

All tissue samples were analyzed by ¹H-NMR spectroscopy at 300 K using a Bruker Avance II 600 spectrometer operating (Bruker Biospin, Germany) at 600.13 MHz. A one-dimensional spectrum was acquired by using a standard (1D) Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence to suppress the water signal with a relaxation delay of 5 sec. Sixty-four free induction decays (FIDs) were collected into 64 K data points with a spectral width of 12,335.5-Hz spectral, an acquisition time of 2.66 sec, and a total pulse recycle delay of 7.66 sec. The FIDs were weighted by a Gaussian function with line-broadening factor -0.3 Hz, Gaussian maximum position 0.1, prior to Fourier transformation [37].

The raw NMR data has been manually Fourier transformed using MestReNova-6.1.1-6384 software before data processing. All of the ¹H NMR spectra were corrected for phase and baseline distortions using MestReNova-6.1.1-6384 software. ¹H NMR spectra of tissue samples were referenced to the TSP resonance at δ0.0. The spectrum ranging from 9.5 to 0.5 ppm was divided into 4500 integral segments of equal length (0.002 ppm). The area under the spectrum was calculated for each segmented region and expressed as an integral value. The region 4.9–4.6 ppm was removed for excluding the effect of imperfect water signal. Moreover, the integrated data were normalized before multivariate statistical analysis to eliminate the dilution or bulk mass differences among samples due to the different weight of tissue, and to give the same total integration value for each spectrum.

For multivariate statistical analysis, the normalized NMR data were imported into SIMCA-P + 11 (Umetrics, AB). The principal component analysis (PCA) was initially applied to analyze the NMR spectral data to separate the tumor samples from the normal samples. The data were visualized using the principal component (PC) score plots to identify general trends and outliers. Orthogonal projection to latent structure with discriminant analysis (OPLS-DA) was subsequently used to improve the separation. R² and Q² values were used to assess the amount of variation represented by the principal components and robustness of the model, respectively. The PLS-DA models were cross-validated by a permutation analysis (200 times) [38, 39]. The default 7-round cross-validation was applied with 1/seventh of the samples being excluded from the mathematical model in each round, in order to guard against over fitting. The model coefficients locate the NMR variables associated with specific interventions as y variables. The model coefficients were then back-calculated from the coefficients incorporating the weight of the variables in order to enhance interpretability of the model: in the coefficient plot, the intensity corresponds to the mean-centered model (variance) and the color-scale derives from the unit variance-scaled model (correlation). The coefficient plots were generated with Matlab scripts with some in-house modifications and were color-coded with the absolute value of coefficients (r) [40, 41].

To identify the variables contributed to the assignment of spectra between tumor tissues and normal controls, the variable importance in the projection (VIP) values of all peaks from OPLS-DA models was analyzed, and variables with VIP > 1 were considered relevant for group discrimination. Moreover, unpaired Student’s t-test (p < 0.05) to the chemical shifts was also used to assess the significance of each metabolite. Only both VIP > 1 of multivariate and p < 0.05 of univariate statistical significance were identified as distinguishing metabolites. The corresponding chemical shift and multiplicity of the metabolites were identified by comparisons with the previous literatures and the Human Metabolome Database (http://​www.​hmdb.​ca/​), a web-based bioinformatics/cheminformatics resource with detailed information about metabolites and metabolic enzymes.