Discrimination of metabolic profiles of pancreatic cancer from chronic pancreatitis by high‐resolution magic angle spinning ¹H nuclear magnetic resonance and principal components analysis

Abstract

The metabolic profiles of Sprague–Dawley rat pancreases were investigated by high‐resolution magic angle spinning proton magnetic resonance spectroscopy (¹H NMR) combined with principal components analysis (PCA) to discriminate pancreatic cancer from chronic pancreatitis. Intact pancreatic tissue samples were obtained from Sprague–Dawley rats with histologically proven pancreatic cancer (n = 5), chronic pancreatitis (n = 5), and two matched controls (n = 5 per group). Two ¹H NMR experiments, single‐pulse and Carr–Purcell–Meiboom–Gill, were carried out separately. Increases in phosphocholine and glycerophosphocholine levels and decreases in leucine, isoleucine, valine, lactate and alanine levels were observed in chronic pancreatitis, whereas the opposite trends were observed in pancreatic cancer. Increasing taurine and decreasing betaine were found both in chronic pancreatitis and in pancreatic cancer. Additionally, the lipid content in pancreatic cancer was higher than that in chronic pancreatitis. PCA was carried out for the single‐pulse and Carr–Purcell–Meiboom–Gill ¹H NMR spectra, respectively, to visualize separation among the samples and to extract characteristic metabolites of pancreatic cancer and chronic pancreatitis. Decreased phosphocholine and glycerophosphocholine were suggested as unique metabolite indicators of pancreatic cancer. Furthermore, even with the disturbance of various quantities of lipid contents pancreatic cancer and chronic pancreatitis could be differentiated well by the combination of high‐resolution magic angle spinning ¹H NMR and PCA. Thus this combination was demonstrated to have the potential to improve magnetic resonance spectroscopy for positive early diagnosis of pancreatic cancer in clinical settings. (Cancer Sci 2007; 98: 1678–1682)

Abbreviations: 

Ala

alanine

Bet

betaine

CCM

choline‐containing metabolites (Cho + PC + GPC)

CDP‐Cho

cytidyl‐diphosphate‐choline Cho, choline

COSY

correlation spectroscopy

CPMG

Carr–Purcell–Meiboom–Gill

CP

chronic pancreatitis

Cr

creatine

DBTC

dibutyltin dichloride

DMBA

dimethyl benzanthracene

FID

free induced decay

GPC

glycerophosphocholine

HRMAS

high‐resolution magic angle spinning

Ileu

iso‐leucine

Lac

lactate

Leu

leucine

LMW

low‐molecular‐weight

MAS

magic angle spinning

MRIS

magnetic resonance imaging spectroscopy

MRS

magnetic resonance spectroscopy

NMR

nuclear magnetic resonance

NS

number of scans

PC

phosphocholine

PCA

principal components analysis

p.p.m.

parts per million

SW

spectral width

Tau

taurine

TD

data points

TE

echo time

TOCSY

total correlation spectroscopy

Val

valine

The pancreas is an integral organ of the digestive system situated deep in the abdominal cavity, sandwiched between the stomach and the spine. This unique location results in unnoticeable signs or symptoms of early stage pancreatic cancer until it has spread to nearby organs such as the stomach, duodenum, liver or gallbladder. Pancreatic cancer metastasizes aggressively even when primary lesions are very minor, forcing the patients to succumb to the disease despite successful surgical removal of the primary cancer. MRIS is a valuable and accurate means for measuring and imaging metabolic processes in vivo, and is also indispensable in confirming and locating pancreatic cancer in clinical diagnosis. However, the most troublesome problem in MRIS diagnosis has always been to differentiate pancreatic cancer from CP and vice versa. Numerous efforts have been made to solve this problem. In immunohistochemical studies Maire and coworkers proved that pancreatic cancer can be excluded with almost certainty with control serum carbohydrate antigen (CA)19.9 levels and the presence of KRAS2 mutation.^{( 1 )} In MRS studies, lipid content has been evaluated to differentiate these two diseases.^{( 2 )} Despite the improvements that have been made in recent years to the MRS instruments, metabolic molecular information is still limited due to the low magnetic field and sensitivity of MRS. The recent combination of HRMAS NMR and MRS provides an exciting method with which to study various tissue abnormalities non‐invasively.

HRMAS ¹H NMR spectroscopy is a well‐recognized technique in metabonomics studies in vitro, by which biopsy or postmortem samples of intact tissues are spun at the magic angle (θ = 54.7°, where θ is the angle between the sample spinning axis and the external magnetic field), resulting in a significant improvement in the resolution of the spectrum obtained for some of line‐broadening factors such as dipole–dipole interactions and chemical shift anisotropy, and magnetic field inhomogeneities are averaged out.^{( 3} , ^{4 )} This approach requires minimal sample preparation and, unlike convenient ¹H NMR spectroscopy of tissue extracts, both aqueous and lipid‐soluble metabolites can be observed simultaneously in situ. In addition, information about the metabolic environment of the tumor can also be obtained. Therefore, HRMAS NMR has proved to be an efficient method for studying of a wide variety of cancers, including breast cancer,^{( 5 )} cervical cancer,^{( 6 )} kidney cancer,^{( 7 )} prostate cancer,^{( 8 )} malignant lymph nodes^{( 9 )} and liposarcoma^{( 10 )} of animals and humans.

HRMAS NMR spectra obtained from tissues reflect the dynamic biological systems and processes that contribute to the overall metabolic status of an organism. It is not possible to isolate the effects of any single metabolite signal in a spectrum and, furthermore, the manual analysis of even a small number of such spectra is a laborious and complex task. Therefore, metabonomists utilize data reduction and multivariate analysis techniques such as PCA to facilitate automated NMR pattern recognition.^{( 11} , ^{12 )} In the present study we applied PCA to the HRMAS ¹H NMR spectra to complete pattern recognition. HRMAS ¹H NMR and PCA were combined to highlight metabolite profiles of pancreatic cancer and CP by which to make a positive diagnosis of pancreatic cancer at its early stage and to discriminate pancreatic cancer from CP accurately without the disturbance of various lipid contents.

Materials and Methods

Animals.  Twenty male Sprague–Dawley rats (approximately 2 months old, weighing 180–200 g, obtained from the animal center of the Second Military Medical University, Shanghai, China) were maintained on standard laboratory feed and water intake in a well‐ventilated room at a temperature of 22 ± 2°C and a relative humidity of 50 ± 10%, with 12:12 h light : dark cycle. Sprague–Dawley rats were divided into four groups: control group I (n = 5), control group II (n = 5), CP group (n = 5) and cancer group (n = 5). The disease model of CP was implemented by a single dose of DBTC solution (intravenous, 8 mg/kg). The DBTC solution was composed of DBTC (Sigma, St Louis, MO, USA), alcohol and glycerol (the molar ratio of DBTC : alcohol : glycerol was 1:2:3). Alternately, an equivalent dose of alcohol and glycerol (alcohol : glycerol, 2:3) was exposed to control group I. The disease model of pancreatic cancer was implemented by implant of 2 mg 7,12‐DMBA (Sigma) in the pancreas. Alternately, 2 mL saline was used in place of DMBA in control group II.

Tissue sampling.  Two months after injection of DBTC, the CP group and control group I (4 months old, weighing 290–320 g) were killed. Ten months after implant of DMBA, the cancer group and control group II (12 months old, weighing ~430–500 g) were killed. The pancreatic tissues were excised from the visible areas of tumors (diameter 1–3 cm) in the cancer group, or from the center parts in the control groups and the CP group. Each tissue sample was cut into two parts: one was taken for histological tests; the other was snap‐frozen in liquid nitrogen and stored at –80°C for running the NMR measurements. Additionally, the surrounding organs close to the tumors in the cancer group were also collected for histological tests.

Histology.  Tissue samples were fixed in formalin, embedded in paraffin and processed routinely. Sections of 4‐µm thickness from all of the groups were stained with hematoxylin–eosin. Those from the CP group and control group I were stained with Van Gieson collagen fiber and those from the cancer group and control group II underwent immunohistochemical examination. Sections were reviewed by two pathologists. It was observed that the lobular structure was destroyed, acini disappeared, collagen fiber proliferated and inflammatory cells were infiltrated in the CP group. Moderately differentiated adenocarcinomas were formed whereas no incidence of adjacent viscera invasion and distant metastasis occurred in the cancer group. No obvious pathological changes in pancreatic tissue occurred either in control group I or control group II. Furthermore, both P53 protein expression and CA19.9 values were positive in the cancer group but negative in control group II.

NMR spectroscopy.  HRMAS ¹H NMR experiments were carried out using a DRX‐500 spectrometer (¹H frequency at 500.13 MHz; Bruker Biospin, Rheinstetten, Germany). Tissue samples were rinsed three times with D₂O and placed into a 4‐mm zirconium oxide MAS rotor with drops of D₂O (deuterium lock reference). Spectra were acquired at 300.0K using single‐pulse and CPMG pulse sequences, both with water presaturation during the relaxation delay of 2 s. CPMG pulse sequence was applied as a T2 filter to suppress signals from the molecules with short T2 values (such as macromolecules and lipids) using a total TE of 320 ms. The main parameters used for ¹H NMR spectra were: SW = 15 kHz; TD = 64 k; NS = 256; and MAS rate = 5 kHz. Spectral assignments were confirmed by 2‐dimensional ¹H–¹H TOCSY and ¹H–¹H COSY (data not shown) together with values obtained from the literature.^{( 3} , ^{13 )}

The stability of tissue samples was evaluated by repeating a 1‐dimensional NMR experiment after overall acquisition. No biochemical degradation was observed for any of the tissue samples.

Principal components analysis.  Spectral data were phased and baseline‐corrected using XWINNMR (Bruker Biospin). All FID were multiplied by an exponential function equivalent to a 0.3‐Hz line broadening factor prior to Fourier transformation. Each HRMAS ¹H NMR spectrum was segmented into 211 regions of equal width (0.04 p.p.m.) over the region δ0.00–10.00 and the signal intensity in each region was integrated using AMIX version 3.6 (Bruker Biospin). The region δ4.50–5.00 was removed to eliminate baseline effects of imperfect water saturation. Prior to PCA, each integral region was normalized by dividing by the sum of all integral regions for each spectrum.^{( 12} , ^{14 )} In order to exclude the effects of lipids and concentrate on the impacts of LMW metabolites in the CCM region, PCA was again done for ¹H CPMG NMR spectra over the range δ2.94–4.40, reducing to 37 regions, each 0.04 p.p.m. wide. PCA was used to calculate a new, smaller set of orthogonal variables from linear combinations of the intensity variables while retaining the maximum variability present within the data. These new variables are the derived principal components, and the distribution of their values (scores) permits the simple visualization of separation or clustering between samples. The weightings (loadings) given to each integral region in calculating the principal components allows for the identification of those spectral regions of greatest influence to the separation and clustering and, hence, the deduction of biomarkers of pancreatic cancer and CP.

Results

HRMAS ¹H NMR spectroscopy.  Fig. 1 shows representative spectra for ¹H NMR of control group I, the CP group and the cancer group. Compared to Fig. 1a, some peaks in Fig. 1b and Fig. 1c present similar aptitudes. For instance, increasing Tau and decreasing Bet were seen in both the CP group and the cancer group despite a more remarkable fluctuation in the CP group. However, the other peaks revealed a distinctively opposite trend: Leu, Ileu, Val, Lac and Ala decreased in the CP group but increased in the cancer group; PC and GPC increased in the CP group but decreased in the cancer group. In addition, the cancer group showed higher lipid levels than the CP group.

Figure 1.

The 500 MHz ¹H high‐resolution magic angle spinning nuclear magnetic resonance Carr–Purcell–Meiboom–Gill spectra of rat pancreases. (a) Control group I; (b) chronic pancreatitis; and (c) pancreatic cancer. For peak assignments, see list of abbreviations used.

Principal components analysis.  In the scores plot based on the single‐pulse NMR spectra (δ0.00–10.00) (Fig. 2a), each group showed its own individual in‐group similarity. The CP group was close to control group I and away from control group II in PC1; the cancer group was far away from control group I and the CP group in both PC1 and PC2 and away from control group II only in PC2. The corresponding loadings plot (Fig. 2b) revealed that it was lipids (δ0.88–0.92 and δ1.28–1.32), Lac (δ1.32–1.36) and CCM, Tau and BET (δ3.20–3.24 and δ3.24–3.28) that contributed most to the separations in Fig. 2a. In order to weaken the effects of lipids and concentrate on the impacts of LMW metabolites, PCA was applied to the whole CPMG NMR spectra range (δ0.00–10.00), obtaining a scores plot and a loadings plot (Fig. 2c,d). In Fig. 2c, all four of the groups were distinguished well. Each group was in its own cluster and positioned apart from the others, including control group I and control group II.

Figure 2.

(a,c,e) Principal components analysis scores plots and (b,d,f) loadings plots. (a,b) From data of the single‐pulse ¹H nuclear magnetic resonance (NMR) spectra (δ0.00–10.00); (c,d) from data of the ¹H Carr–Purcell–Meiboom–Gill (CPMG) NMR spectra (δ0.00–10.00); (e,f) from data of the ¹H CPMG NMR spectra (δ2.94–4.40) (○, control group I; +, control group II; Δ, chronic pancreatitis; ×, pancreatic cancer).

It was reported that proliferation of cancer cells is associated with CCM. However, there has been no related publication about pancreatic cancer. To straighten out the relationship between CCM and pancreatic disease, PCA was applied again to ¹H CPMG NMR spectra over the range δ2.94–4.40, resulting in a scores plot and a loadings plot (Fig. 2e,f). In Fig. 2e control group I and control group II congregated. Whether in PC1 or PC2, the control, pancreatitis and cancer groups stood apart from each other.

Discussion

Lactate, CCM and various lipids measured with HRMAS NMR proved to be correlated with cancer.^{( 15} , ¹⁶ , ^{17 )} In our studies, pancreatic cancer showed higher Lac, lipid and Tau levels and lower Bet, PC and GPC levels relative to CP. These results suggest that lipids, lactate and metabolites in the CCM region can be taken as metabolic markers for differentiating pancreatic cancer from CP.

Lactate, the product of anaerobic glycolysis, is increased in hypoxia, ischemia and poorly vascularized cancer.^{( 18 )} It was also demonstrated by Terpstra et al. that the lactate pool was metabolically active in the majority of the cancer.^{( 19 )} Our results showed that Lac was increased highly in pancreatic cancer and decreased slightly in CP due to the association of Lac with metastatic potential as reported by Sitter et al.^{( 6 )} Compared to the other LMW metabolites, Lac had the most noticeable difference between CP and pancreatic cancer.

Choline‐containing metabolites have already been chosen as biomarkers in various carcinoma studies,^{( 17} , ^{20 )} yet have not been mentioned in pancreatic cancer so far. Noticeably, CCM were increased in most cancer tissues, which was explained as a result of high membrane concentrations during high proliferation of cancer cells. However, in our results PC and GPC decreased in pancreatic cancer but increased in CP. As the modest MAS rates (<6 kHz) are less than the dipole–dipole interactions of the cell membrane,^{( 21} , ²² , ^{23 )} membrane‐bound CCM are expected to be poorly observed. Hence, CCM detected in our study mainly derived from free molecules in the pancreas. The unchanged Cho and decreased PC and GPC in pancreatic cancer may result from blockage of Cho‐kinase and PC transferase or consumption of PC through the CDP‐Cho pathway.^{( 24} , ^{25 )} Thus, we may deduce that decreasing PC and GPC could be taken as a unique profile of pancreatic cancer.

Elevated Tau and depleted Bet also played appreciable roles in the discrimination of pancreatic cancer from CP. Tau is a principal free intracellular amino acid and its elevation has been investigated in squamous‐cell carcinoma,^{( 26 )} prostate cancer, liver metastasis^{( 27 )} and malignant breast tissues.^{( 28 )} Because the metabolic actions of Tau were correlated with control of serum cholesterol levels, blood sugar and cellular calcium levels, Tau tests by HRMAS NMR might be combined with blood tests such as CA19.9 levels. Bet donates methyl groups for remethylation of homocysteine to methionine and dimethylglycine, which support proper liver and pancreatic function, cellular replication and detoxification reactions. Because Cho is the precursor of Bet, the depletion of both Bet‐ and Cho‐containing metabolites in pancreatic cancer must be interrelated.

Metabolites around δ3.20 in the CCM region, including Cho, PC, GPC, Tau and Bet, investigated by HRMAS ¹H NMR spectroscopy, can be seen individually. However, they always form a hump in MRS spectra, which complicates the spectral interpretation. Thus, HRMAS NMR in vitro compensates for MRS in vivo.

Clinical research has indicated that high lipid content can be used to differentiate pancreatic cancer from CP.^{( 2 )} Cancerous tissues generally have high lipid content but the reverse is not always true for non‐cancerous tissue because lipid contents vary widely due to influences such as genetics, aging, sex and dietary variation. In the HRMAS ¹H NMR spectra we observed that the lipid levels of the cancer group were higher than those of the CP group, whereas the lipid levels of control group II (12 months old) were also higher than those of control group I (4 months old). The different lipid levels between the two control groups were mainly due to the age difference. It was this difference that brought out their separation in the scores plot from single‐pulse NMR spectra without suppression of lipid signals (Fig. 2a). This observation gave us a hint that the different lipid levels between the CP and cancer groups might be caused by age differences as well as the different disease types.

Moreover, the scores plot (Fig. 2a) showed that the CP group was mixed with control group I and the cancer group could not be separated from control group II in PC1. This was due to the high weightings of lipids because the loadings plot (Fig. 2b) showed that almost one‐half of the regions corresponded to lipids. In order to weaken the effects of lipids the CPMG NMR experiments were carried out, to which PCA was applied. As a result the CP and control I groups were separated distinctly (Fig. 2c). However, the separations among the other groups didn't have perceptible improvements because the residual lipid peaks of the cancer group and control group II were still weighted heavily (shown in Fig. 2d). It is obvious in Fig. 2b,d that besides lipids the weighted regions came mostly from the CCM. This prompted us to exclude the lipid effects and use PCA only over the CCM region (δ2.94–4.40), which contains the majority of LMW metabolites relative to pathophysiological processes. Figure 2e was then obtained where the control groups, CP group and cancer group presented the best in‐group similarity and the best separations on both PC1 and PC2. Furthermore, control II group migrated toward control I and merged completely together due to their similar inherent metabolic profiles, except for the different lipid contents. However, the cancer and pancreatic groups were always separated well in all of the scores plots as shown in Fig. 2a,c,e due to their varying lipid contents and different disease types.

Although the number of samples in our study was limited the potential of HRMAS NMR for the in vitro investigation of pancreatic diseases should not be ignored. The above results clearly demonstrate that the metabolic profiles of pancreatic cancer and CP can be discriminated characteristically without the disturbance of various lipid contents by HRMAS ¹H NMR coupled with PCA. Because metabolite changes observed by HRMAS NMR always occur before morphological changes investigated by MRIS, HRMAS NMR will certainly benefit pathological research, early diagnosis and therapy evaluation of pancreatic diseases.