Non-invasively predicting differentiation of pancreatic cancer through comparative serum metabonomic profiling

PDAC cell strains (Panc-1, BxPC-3 and SW1990, Catalog NO. SCSP-535, TCHu 12 and TCHu 201) were obtained from Shanghai Institute of Cell Biology, Chinese Academy of Sciences (Shanghai, China) authenticated with short tandem repeat test and mycoplasma culture. At the circumstance of 5% CO₂ and 37 °C, these strains were incubated in dulbecco’s modified eagle medium (DMEM, Gibco, Thermo Fisher Scientific Inc., USA) added with 10% fetal bovine serum (Gibco) in cell incubator (3110, Thermo Scientific). Then, cells were digested by 0.125% trypsinogen (Life Technologies, Grand Island, NY, USA) for the passage with the ratio of 1:2-4 every 2-3 days. BALB/c nude mice (male, 4 weeks, weighing 18-20 g), purchased from Shanghai Slac laboratory animals Co., Ltd. (NO: SCXK (HU) 2012-0002), were bred in Fujian Medical University Animals Centre (Fuzhou, china) with a standard SPF-grade laboratory conditions.

This experimental protocol was in accordance with the principles of National Institutes of Health guide for the care and use of laboratory animals and approved by Ethical Committee of Fujian Medical University. Three PDAC cell strains in the exponential phase were digested with 0.125% trypsinogen, washed by phosphate buffered saline (PBS) for three times, then collected and resuspended in PBS (1 × 10⁷ cells per milliliter). After skin degerming, the cell suspension liquids were subcutaneously injected into the axilla of mice (one cell strain each mouse), followed by a month of normal feeding. The tumors with a size of 5 to 10 mm in diameter generated in the injected positions of mice. Consequently, the mice were executed by a mercy killing, and the tumor tissues of Panc-1, BxPC-3 and SW1990 were carefully collected and divided into pieces of 1 mm³ for implantation in situ.

Thirty days after surgeries, 1 mL of blood from each group was collected by aortic puncture under continuous airway anesthesia of isoflurane (Jiupai pharmaceutical Co., Ltd., Shijiazhuang, China) and stored in clear 1.5-mL Eppendorf tubes. After standing for 30 min, the blood went through a 10-min centrifugation at 10,000 g and 4 °C. The supernate was collected and immediately frozen by liquid nitrogen and stored at −80 °C. For the detection of ¹H NMR spectroscopy, 400 μL of serum were melted on the surface of ice, and then mixed with 200 μL of 90 mM deuterated phosphate buffer (NaH₂PO₄ and K₂HPO₄, pH 7.4). The mixture of serum and buffer were centrifuged again, and finally, 550 μL of the supernate was moved into 5-mm NMR tubes (ST500, NORELL, Inc., Morganton, North Carolina, USA).

The ¹H NMR spectroscopy of serum samples were performed on a Varian NMR system (Agilent Technologies Co, Palo Alto, California, USA) with a 500.13 MHz of proton frequency at the temperature of 298 K. For each sample, a water-suppressed CPMG (Carr-Purcell-Meiboom-Gill) spin-echo pulse sequence (RD-90°-(τ-180°-τ)_n-ACQ) was used to acquire the NMR spectrum. Herein, a total of 64 scans with a spectral width of 6 KHz and a data point of 12 K were accumulated for all spectra. Spin-echo loop time (2nτ) of 70 ms was applied with a relaxation delay of 2.0 s. The NMR spectra were processed by using MestReNova (V9.0.1, Mestrelab Research S. L., Spain). In order to increase the signal-to-noise ratio, all free induction decays were multiplied by an exponential weighting function equivalent to a 1 Hz line-broadening and subsequently disposed with Fourier transformation. To make the spectra more comparable, the manual phase rectifications and baseline corrections were conducted by using MestReNova. The chemical shifts were referenced to the double-peak of endogenic lactate at δ1.33 for metabolites identification. Automatically, the spectral regions δ9.0-0.5 of the processed NMR spectra were segmented into scatter integral regions of 0.002 ppm with a removal of spectral region δ6.40-5.50 and δ5.19-4.36 to eliminate the impacts of residual water signal and urea signal, respectively. Finally, for each spectrum, the integrated data were normalized to the total sum of the spectrum in favour of multivariate statistical analysis.

The multivariate statistical analysis, including principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA) and orthogonal partial least squares discriminant analysis (OPLS-DA), were performed in SIMCA-P⁺ (V14.0 Umetrics, Sweden) to analyze the metabonomic differences between three PDAC groups. PCA, performed in the approach of mean-centered scaling, could simplify the normalized date into several components, which can roughly evaluate the clusters distributions and identify the existence of outlines. PLS-DA and OPLS-DA, which can be classified as supervised multivariate statistical analysis, were conducted in the approach of parato-scaling approach for better extraction and maximization of the metabonomic differences between PDAC groups. Furthermore, the OPLS-DA models coefficients, which were back-calculated from the coefficients, incorporated with the weight of the variables, and then to be plotted with color-coded coefficients to enhance interpretability of the models. As a result, the metabolites responsible for the metabonomic differences between groups can be extracted from the corresponding color-coded loading plots and displayed visually. By assistance of MATLAB (V7.1, the Mathworks Inc., Natick, USA), the color-coded coefficient loading plots were drew and color-coded according to the absolute value of coefficient. That meant, in the loading plots, a warm-toned color (i.e. red) represents for the metabolites being positive or negative significant in distinguishing different groups while a cool-toned color (i.e. blue) corresponds to the metabolites not being significant in discriminations. Moreover, to screen out differential metabolites, the cutoff value of correlation coefficients (|r| > 0.576) was determined according to the statistical significance of the Pearson correlation coefficient test at the level of P < 0.05 and df (degree of freedom) =10. In order to assess the quality and validity of models, the 10-fold cross validation and response permutation testing (n = 200) were performed and the corresponding parameters R² and Q² in the permutated plots presented the degree of model fitting and the potentially predictive ability of models, respectively.

After visual confirmation for tumorgenesis, 12, 13, and 11 serum samples from Panc-1, BxPC-3 and SW1990 groups were included for the detections with ¹H NMR spectroscopy, respectively. Typical one-dimensional 500-MHz ¹H NMR spectra of serum samples from models induced by the different differentiated PDAC cells are presented in Fig. 1, which provided an integrated overview of all metabolites. Forty-seven metabolites were identified from the NMR spectra (Table 1) based on the relative literatures and public databases [42, 43]. A certain degree of metabolic differences could be noticed between different PDAC groups visually such as ethanol and phosphocholine. But considering the high similarity of spectra, the metabonomic information acquired was quite limited and the multivariate statistic analysis will help to extract the detailed information.

To show an overview of ¹H NMR data collected from the serum of Panc-1, BxPC-3, and SW1990 groups, the PCA and PLS-DA were performed. The PCA scores plot showed a certain degree of separated trends between the three PDAC groups (Fig. 2a) though a little overlap or dispersity was demonstrated, indicating their obvious metabonomic differences. In further, a greater discrimination in cluster distributions of Panc-1, Bxpc-3 and SW1990 could be observed visually in PLS-DA scores plot (Fig. 2b), demonstrating a significant differences with each other.

To get deep insight into the metabolites responsible for the metabonomic alterations occurred in three PDAC groups, pair-wise comparisons were conducted by using the PLS-DA combined with orthogonal projection (OPLS-DA). The pronounced separations were demonstrated in OPLS-DA scores plots (Fig. 3 upper left panels) and the metabolites corresponding to the metabolic difference were marked in loading plots (Fig. 3 bottom panels). The summarized dominant metabolites, based on the cutoff value of correlation coefficient (|r| > 0.576), and the correlation coefficients were listed in detail based on their biochemical types (Table 2). Overall, the levels of metabolites belonged to glycolysis and glutaminolysis, alcohols and amino acids were lower in SW1990 group while the high concentrations of choline and its derivatives were noticeable in Panc-1 group. The favorable fit and prediction parameters (R² and Q²) of the OPLS-DA models and the corresponding permutation test and probability (p-value) via CV-ANOVA also confirmed the strong predictive ability of the models to guarantee a reliable identification of characteristic metabolites.

For better understanding of the bioinformation contained in discriminatory metabolites, the biochemical pathways were identified based on the differential metabolites derived from OPLS-DA of pair-comparisons and those with p-value less than 0.01 were demonstrated on Fig. 4. The p-value for pathway identification were calculated automatically by the MBROLE [40].

In the analysis to compare SW1990 with Bxpc-3, the numerous amino acid-related pathways were noticeable, including metabolism of essential and non-essential amino acids, the biosynthesis of aminoacyl-tRNA and ABC transporters. In addition, the pathways related with glycolysis involving pyruvate, galactose, glutamine and glutamate were also identified as differential features to distinguish the Bxpc-3 from the SW1990. Meanwhile, except the pathways of lysine, histidine and thiamine metabolisms, most pathways involved in Bxpc-3 vs SW1990 were also identified in the comparison between Panc-1 and Sw1990. In addition, the pathways of glycerophospholipid metabolism and the degradation of valine, leucine and isoleucine were also identified to be a signature contributed to distinguish Panc-1 from SW1990. In term of metabolic diversity between Panc-1 and BxPC-3, the metabolic discrimination seems to be quite limited where only a few pathways related with amino acids and glycerophospholipid metabolism were identified.

In present study, we found that the high concentration of citrate, lactate, glutamate and glutamine can help to distinguish the SW1990 from Panc-1 and Bxpc-3. Being well known, the tumor metabolic reprogramming has been validated to be the cornerstone for malignant transformation and one common composition in this process is the aerobic glycolysis (Warburg effect). Through the aerobic glycolysis, rather than tricarboxylic acid (TCA) cycle, the tumor cells derive the predominant ATP/energy and generate extensive lactate from pyruvate to result in environmental acidosis which promote the spreading of the tumor cells [44]. Meanwhile, the lactate generated from hypoxic PDAC can be taken up by normoxic PDAC cells nearby as fuel to maintain proliferation, creating a phenomenon called tumor symbiosis [45]. Thus, the tumor metabolic impact upon the level of lactate in peripheral circulation may be determined by the dynamic balance of release and uptake of lactate around tumor microenvironment. Our outcome indicates that the tumor with a poorer differentiation could induce a lower concentration of lactate in serum relative to that with a better differentiation, which may be due to a stronger ability of lactate recirculation. It’s also implied by inconsistent variation trends of lactate in serum reported by previous studies [46, 47]. In addition, due to the breakdown of TCA cycle, glutaminolysis is enhanced in PDAC cells to generate TCA intermediates (e.g. malate, oxaloacetate and citrate) which is called anaplerosis reaction, and subsequently served as building blocks for synthesis of lipid and non-essential amino acids [48]. Besides, glutamine can also act as fuel to support energy metabolism through aspartate, oxaloacetate and pyruvate transformation process, thus promoting growth of pancreatic cancer via Kras-regulated metabolic pathway [49]. Therefore, the significantly low levels of glutamine, glutamate and citrate may indicate that the tumor with poorer differentiation may provide a more dramatic glutaminolysis and deprive more glutamine and glutamate from peripheral circulation.

Likewise, the higher concentrations of amino acids could also contribute to the distinguishing of the SW1990 from Panc-1 and Bxpc-3, which could serve as key participants in the cancer metabolism reprogramming. Under the influence of the abnormal expression of oncogenes and tumor suppress genes, the anabolic metabolism and transport of amino acid were tremendously enhanced for rapid proliferation of cancer cells. To provide required nutrients for cancer growth, the catabolic metabolism of whole-body tissue would be enhanced, leading to an increased circulating amino acids at the early stage of PDAC [50]. But the catabolic metabolism cannot maintain in a high level for a long time and end in a severe nutritional imbalance called cachexia, thus creating a decrease of amino acids in serum at last. In this process, L-type amino-acid transporter 1 (LAT-1), the most important transporter of neutral amino acids, plays a key role in internalized transportation of essential amino acids (EAAs) in PDAC. As previous reports demonstrated, the overexpression of LAT-1 can promote cancer growth via mammalian target-of-rapamycin (mTOR) and serve as a prognostic factor in PDAC [51, 52]. Thus, the higher concentration of EAAs in SW1990 group than in Panc-1 and BxPC-3 group indicates that the tumors with poor differentiation may have a higher expression of LAT1 and nutritional stress from rapid proliferation, which can associated with poor prognosis.

With regard to the non-essential amino acids (NEAAs), several pathways were involved to enhance their biosynthesis and utilization for cell proliferation. As noted above, the accumulated glycolysis intermediates could also promote the biosynthesis of glycine, serine and threonine through 3-phospho-D-glycerate pathway. In addition, the increased glutaminolysis provides numerous substrates (e.g. isocitrate, malate, alpha-ketoglutaric acid) not only to supply the lipids synthesis but also to promote the biosynthesis of alanine and aspartate. Besides being used as building blocks and fuels for cell proliferation, NEAAs have been indicated to bridge the interplay metabolism and epigenetics, thus serve as programmed switch for cell differentiation [53]. For instance, several NEAAs including glycine could be associated with gene signatures of cell proliferation and Myc target activation through the serine-glycine-one-carbon pathway (SGOC pathway), which contribute significantly to energy generation and biosynthesis of NADPH and purine [54]. In addition, the mTOR-dependent induction of SGOC pathways can also lead to DNA methylation and tumorigenesis under the cooperatively oncogenic function of the loss of liver kinase B1 and activation of Kras, which highly involved in epigenetics [55]. Thus, NEAAs are highly associated with genesis, progression and epigenetics, and their relative concentration in serum may be indicators for the differentiation of PDAC.

Impressively, the high correlation coefficient of choline groups in the pair-comparison of Panc-1 vs BxPC-3 and Panc-1 vs SW1990 implied that relatively high concentration of choline-like metabolites including phosphocholine (PC) and glycerolphosphocholine (GPC) may be significant metabolic features for poor differentiation of PDAC. According to previous study, the tumor-associated choline metabolism plays a key role in cell malignant transformation, tumor migration and metastasis [56, 57], characterized by elevated level of PC and total choline in tissue [45, 46]. Thereinto, the overexpression of choline kinase-α (Chk-a) induced by hypoxia-inducible factor (HIF) accounts for the increase of cellular PC and total choline [58], generating excessive phosphatidylcholine for biosynthesis of cell membrane. In addition, the EDI3-intermediated choline metabolism, a pathways verified in other solid tumor, can not only cleave GPC to form choline to supplement Kennedy pathway, but also generate glycerol-3-phosphate and its sequentially downstream intermediators for cellular signaling to regulate migration, invasion, proliferation and differentiation [57]. Other research detecting serum from PDAC patients also indicated that the choline metabolism were obviously altered and could potentially serve as biomarkers to detect PDAC in early stage. Thus, the difference of choline metabolism in PDAC could reflect and create a handful of regulatory functions on tumor progression and differentiation.

There are some pitfalls in this study. The PDAC models were established by using three PDAC cell lines which could only represent a part of metabonome landscape of pancreatic cancer, which inevitably lower the level of evidence provided from our study. The heterogeneity of pancreatic cancer in patients may compromise the directly clinical transformation application of our results. Thus, the further validation based on a large patient cohort will be performed in the future.