Appl Magn Reson (2014) 45:19–35 DOI 10.1007/s00723-013-0501-7
Applied Magnetic Resonance
Differentiation Between Hepatocellular Carcinoma and Colorectal Cancer Liver Metastases on HighResolution Magic Angle Spinning Spectroscopy: Preliminary Study So Yeon Kim • Siwon Kim • Chul-Woong Woo • Jae Ho Byun Seung Soo Lee • Moon-Gyu Lee • Haeryoung Kim • Kyoung Ho Lee • Young Hoon Kim • Jai Young Cho • Suhkmann Kim • Jin Seong Lee
•
Received: 11 July 2013 / Revised: 17 October 2013 / Published online: 8 December 2013 Ó Springer-Verlag Wien 2013
Abstract To explore the potential of high-resolution magic angle spinning (HRMAS) 1H nuclear magnetic resonance (NMR) spectroscopy for differentiation and metabolite characterization of hepatocellular carcinoma (HCC) and colorectal liver metastases (CRLM), we prospectively included 21 pathologically confirmed malignant hepatic tumors (8 HCC and 13 CRLM) and 26 non-tumorous hepatic
S. Kim and J. S. Lee contributed equally to this work as corresponding authors. S. Y. Kim J. H. Byun S. S. Lee M.-G. Lee J. S. Lee (&) Department of Radiology and Research Institute of Radiology, Asan Medical Center, University of Ulsan College of Medicine, 388-1, Pungnap-2 dong, Songpa-ku, Seoul 138-736, Korea e-mail:
[email protected] S. Y. Kim K. H. Lee Y. H. Kim Department of Radiology, Seoul National University Bundang Hospital, 166 Gumiro, Bundang-gu, Seongnam-si, Gyeonggi-do 463-707, Korea S. Kim S. Kim (&) Department of Chemistry, Chemistry Institute for Functional Materials, Pusan National University, 30 Jangjeon-dong, Geumjeong-gu, Busan 609-735, Korea e-mail:
[email protected] C.-W. Woo Asan Institute for Life Sciences, Asan Medical Center, Seoul, Korea H. Kim Department of Pathology, Seoul National University Bundang Hospital, Seongnam-si, Gyeonggi-do, Korea J. Y. Cho Department of Surgery, Seoul National University Bundang Hospital, Seongnam-si, Gyeonggi-do, Korea
123
20
S. Y. Kim et al.
parenchyma from 26 patients who underwent hepatic tumor resection. Using intact tissue samples obtained during surgery, HRMAS 1H NMR spectroscopy was performed at 11.7 T. All observable metabolite signals were acquired using a waterpresaturated standard one-dimensional Carr–Purcell–Meiboom–Gill sequence. Metabolomic profiles contributing to the differentiation of HCC and CRLM and of each tumor and non-tumorous hepatic parenchyma were represented by orthogonal partial least squares discriminant analysis (OPLS-DA) and loading plots. Metabolite intensity normalized by total spectral intensities in both tumors was compared using student’s t tests. OPLS-DA and loading plots demonstrated good separation between tumors and non-tumorous hepatic parenchyma. The metabolomic characteristics of HCC showed separation from those of CRLMs according to OPLS-DA. Compared with CRLM, HCC showed significantly elevated levels of glucose (P \ 0.01) and sn-Glycero-3-phosphocholine (P \ 0.01), and decreased levels of hypoxanthine (P = 0.04). HCC and CRLM could be differentiated by the metabolic profile using HRMAS 1H NMR spectroscopy.
1 Introduction Metabolomics is the global quantitative assessment of endogenous metabolites within a biological system. Within the context of the immediate environment, metabolites reflect changes in genes and proteins and, therefore, function as a terminal view of the biological system [1]. As cancer cells are known to possess a highly unique metabolic phenotype according to their specific genetic and proteomic properties, metabolomics can be a potential method for finding a biomarker in cancer diagnosis, prognosis, and therapeutic evaluation [1–3]. Among the various techniques used in metabolomics, high-resolution magicangle spinning (HRMAS) 1H nuclear magnetic resonance (NMR) spectroscopy is a powerful technique for investigating metabolic profiles. As the physical effects, such as dipole–dipole interactions that broaden lines in the solid state, are reduced by spinning a sample at the magic angle (54.7°), HRMAS 1H NMR spectroscopy makes it possible to obtain high-resolution spectra on a sample as small as 45 lL in intact tissue without any specific destructive sample preparation [4]. Recently, HRMAS 1H NMR spectroscopy has been increasingly used to evaluate hepatic tumors and hepatic parenchyma [5–11]. Identification of the different metabolite profiles in primary and metastatic tumors would be helpful for the differential diagnosis between the primary tumor and the metastatic tumors. The two most common hepatic malignant tumors are hepatocellular carcinoma (HCC), which is the most common primary hepatic tumor, and colorectal liver metastases (CRLM), which is the most common metastatic malignancy in the liver [12]. Considering that previous studies have investigated with proteomic approaches to differentiating HCC and CRLM [13], the discriminating metabolic profiles between these tumors are also expected to be present. Even though HRMAS 1H NMR spectroscopy has been used to evaluate the metabolic changes in HCC [10, 11], to our knowledge, there has not been a metabolomic approach to determining the different biomarkers of HCC and CRLM.
123
Differentiation Between Hepatocellular
21
Therefore, the purpose of our study was to explore the potential use of HRMAS H NMR spectroscopy for the differentiation of HCC and CRLM by the metabolic characterization. 1
2 Methods The study protocol was approved by institutional review board of Seoul National University Bundang Hospital, and written informed consent was obtained from all participants. 2.1 Study Subjects Between June 2010 and February 2011, 26 (17 men and eight women; mean age ± SD 57.3 years ± 11.3; age range 33–74 years), consecutive patients who had been referred for hepatic resection for suspected HCC (n = 12) or CRLM (n = 14) noted on preoperative imaging evaluation were prospectively enrolled in our study. We did not include patients who had undergone any specific anti-tumor treatment as this could alter the metabolites of the malignant hepatic lesions [14]. Patients whose tumors were confirmed as cholangiocarcinoma (n = 2) and combined hepatocellular-cholangiocarcinoma (n = 2) on the final pathology reports were excluded from analysis of the tumor spectrum. Principal component analysis (PCA) was first performed on the spectra to offer the possibility of detecting outliers [15–17]. Outliers are found in the PCA score plots by Hotelling’s T2. Since an ellipse in the PCA score plots refers to Hotelling’s T2 as 95 %, we determined data outside the ellipse as outliers. The one CRLM sample was an outlier and inspection of the spectra revealed abnormally high levels of lipid thus implying necrosis as histopathologic examination confirmed the presence of extensive necrosis. Therefore, it was excluded from further analysis. Finally, the remaining 21 patients were included for analysis of the tumor spectrum. The data from all 26 patients were included for analysis of the non-tumorous hepatic parenchyma. 2.2 Collection of Specimens All tissue samples were dissected from specimens at operation room immediately after the partial hepatectomy was performed. Samples from the tumor and nontumorous hepatic parenchyma were obtained separately and were put into separate cryo-tubes. When the tumor samples were obtained, we meticulously attempted not to include either non-tumorous hepatic parenchyma or necrotic tumor area, based on the macroscopic findings. When we obtained a sample of non-tumorous hepatic parenchyma, it was obtained as far as possible from the tumor, based on the macroscopic findings. The samples were snap-frozen in liquid nitrogen and were then stored in a -80 °C freezer.
123
22
S. Y. Kim et al.
2.3 HRMAS 1H NMR Spectroscopy Tissue samples were defrosted at room temperature immediately prior to analysis. One part of a sample (15–25 mg) was used for spectroscopic analysis, and the nearest part to that was fixed in 10 % formalin for pathology examination. Each sample was then placed in a 4-mm-diameter zirconium rotor and D2O was added into the rotor to provide field-lock signal for NMR examination. All of the 1H NMR spectra were recorded at 25 °C measured by the thermocouple system and maintained by the cooling of the inlet gas on an NMR spectrometer (500 MHz Varian Unity-Inova; Varian Inc., Palo Alto, CA, USA) operating at 11.7 T and equipped with a gHX nano probe [18]. The rotation rate was 2.0 kHz to move the spinning side band outside the region of interest of the spectrum. The water signal was suppressed using a water-presaturated, standard, one-dimensional Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence (recyle delay–90–(s–180–s)n–acquisition where s = 60 ms and n = 128) [19]. Signal was acquired after a 90° pulse on 16 K data points for a spectral window of 6,000 Hz. The typical 1H 90° pulse length and relaxation delay were, respectively, 5.0 ls and 1 s [20]. Spectra were measured with 128 transients. These CPMG spectra showed less or no broad signal from the proteins and large molecules when compared with single pulse pre-saturation or to NOESY pre-saturation spectra [21]. 2.4 HRMAS 1H NMR Spectroscopic Data Analysis HRMAS spectroscopic data were processed using Chenomx NMR Suite 6.01 software (Chenomx Inc., Canada) with an exponential function corresponding to a 1 Hz line broadening prior to Fourier transformation. Baseline correction was performed using a three-order polynomial function. The spectral region between d 0.4 and 9 ppm was binned into 835 spectral regions of 0.01-ppm width. The region of d 4.65–4.9 was discarded to eliminate the effect of the residual water signal. The binned data are designated by their central chemical shift value. Their most probable assignment to metabolites was given in reference to the spectral assignment delineated in previous literatures [5, 6, 9, 22] (Table 1). We also took into account the detailed spectral appearance (multiplet, high resolution peaks, and broad signals) and compared it with the width of the 0.01 ppm binned data. The resonance intensities of each metabolite reported here represented integrals of curve-fittings with Lorentzian-Gaussian line shapes of each of the binned data. As absolute concentration quantification for metabolites is known to be difficult in HRMAS spectroscopy, the relative quantification was commonly used for statistical analysis [23, 24]. For the relative quantification of the metabolite concentrations, the integral area under the each metabolic peak of the 33 most intense resonance peas with 0.01 ppm interval was normalized by the sum of all the integral areas prior to performing pattern recognition analysis. During preprocessing spectroscopic data, uninformative regions of the spectra were removed manually, i.e. the water residual in water suppressed spectra or pure baseline regions, based on a priori knowledge. It was performed by an operator (S.K.) blinded to all tissue pathologic information.
123
Differentiation Between Hepatocellular
23
Table 1 HRMAS 1H NMR data for tumors and nontumorous hepatic parenchyma Metabolite
Chemical shifts and peak shape, ppm
3-Hydroxybutyrate
1.20(d), 2.32 (dd), 2.39(dd), 4.14(dt)
Acetate
1.92(s)
Acetoacetate
2.27(s), 3.44(s)
Adipate
1.58 (m), 2.25(m)
Alanine
1.47(d), 3.78(q)
Arginine
1.69(m), 1.90(m), 3.24(t), 3.76(t)
Asparagine
2.85(m), 2.94(m), 4.00(dd)
Aspartate
2.67(m), 2.81(m), 3.89(m)
Choline
3.20(s), 3.51(m), 4.07(m)
Creatine
3.03(s), 3.92(s)
Creatinine
3.04(s), 4.05(s)
Cysteine
3.02(m), 3.10(m), 3.97(m)
Glucose
3.24(dd), 3.40(m), 3.47(m), 3.53(dd), 3.72(m), 3.84(m), 3.90(dd), 4.64(d), 5.23(d)
Glutamate
2.08(m), 2.35(m), 3.75(m)
Glutamine
2.13(m), 2.44(m), 3.77(t)
Glycerol
3.56(m), 3.65(m), 3.78(m)
Glycine
3.56(s)
Hypoxanthine
8.18(s), 8.21(s)
Inosine
3.83(m), 3.91(m), 4.27(m), 4.43(m), 4.77(m), 6.09(d), 8.23(s), 8.34(s)
Isoleucine
0.93(t), 1.01(d), 1.25(m), 1.46(m), 1.97(m), 3.66(d)
Lactate
1.33(d), 4.11(q)
Leucine
0.96(dd), 1.70(m), 3.73(m)
Lysine
1.46(m), 1.72(m), 1.89(m), 3.02(t), 3.75(t)
Methionine
2.11(m), 2.13(s), 2.19(m), 2.63(t), 3.85(m)
myo-Inositol
3.27(t), 3.53(dd), 3.61(t), 4.06(t)
O- Phosphocholine
3.22(s), 3.59(m), 4.17(m)
Phenylalanine
3.11 (m), 3.28(dd), 7.32 (m), 7.37(m), 7.42(m)
sn-Glycero-3phosphocholine
3.22(s), 3.60(m), 3.68(m), 3.91(m), 4.31(m)
Succinate
2.40(s)
Taurine
3.26(t), 3.41(t)
Trimethylamine N-oxide
3.26 (s)
Tyrosine
3.04(dd), 3.19(dd), 3.93(dd), 6.89(d), 7.18(d)
Valine
0.99(d), 1.04(d), 2.26(m), 3.60(d)
d doblet, dd two doublets, dt doublet and triplet, s singlet, m multiplet
2.5 Histopathology All tumor samples and tissue samples were fixed in formalin, embedded in paraffin, and cut into 5-mm sections. The sections were stained using the routine hematoxylin and eosin (H&E) methods. In general, two to five slices were examined for each
123
24
S. Y. Kim et al.
sample by a pathologist (H.K.). According to the results of the pathology examination, tumor samples were classified as HCC, CRLM, and non-tumorous liver sample. The tumor samples were graded according to the World Health Organization (WHO) classification criteria [25]. 2.6 Statistical Analysis Statistical analyses for spectroscopy were performed using SIMCA-P ? software (version 12.0, Umetrics AB, Umea, Sweden). A PCA was performed to provide the possibility for detecting outliers, as described previously. The orthogonal partial least squares discriminant analysis (OPLS-DA) test, a supervised multivariate data analysis method, was used to test the hypothesis that different pathologic components have differing metabolic profiles [15–17, 21, 26]. The ability of the model to describe data and to correctly predict new data was expressed by the value of the parameters R2 and Q2, respectively [17, 21]. R2 and Q2 are quantitative measures of an OPLS-DA model. R2 refers the goodness of fit (i.e. data variation) of a model. On the other hand, Q2 indicates the goodness of prediction, as estimated by cross-validation. Generally, a prediction model is considered to be good when Q2 [ 0.5, and excellent if Q2 [ 0.9 [16, 27]. Results were visualized by scores and loadings plots which contain the information related to class separation. From the score and loading plots, the marker metabolites responsible for the classification were identified in consideration of magnitude and reliability. Reliability was determined by consistent upward or downward direction of each metabolite on loading plots. We also compared the relative concentrations of marker metabolites that contributed to the classification of HCCs and CRLMs as well as the classification of each tumor and the non-tumorous hepatic parenchyma [10, 15, 28]. Since pairwise comparisons were performed using the t tests, P values were expressed as Bonferroni-corrected values. Statistical analyses for comparison of the concentrations were performed using commercial software (PASW, Version 17; SPSS Inc., Chicago, IL, USA; MedCalc, Mariakerke, Belgium). A P value \0.05 was considered significant.
3 Results 3.1 Histopathologic Analysis In total, HRMAS 1H NMR spectroscopic data from 21 tumor samples which included 8 HCC, 13 CRLM, and 26 non-tumorous hepatic parenchyma samples were analyzed. The histopathologic characteristics of the tumors and the nontumorous hepatic parenchyma are summarized in Table 2. All eight HCC included in our study were moderately differentiated HCC. Our study included various histologic grades of CRLM and various conditions of the non-tumorous hepatic parenchyma, such as normal (n = 15), steatosis (n = 1), chronic hepatitis (n = 6), and liver cirrhosis (n = 4).
123
Differentiation Between Hepatocellular
25
Table 2 Characteristics of patients, tumors, and nontumorous hepatic parenchyma Age (years)
Sex Tumor (cause of exclusion)
WHO tumor grade
Nontumorous hepatic parenchyma
66
F
CRLM
PD
Normal
68
M
CRLM
MD
Normal
51
F
Excluded (combined)
NA
Chronic hepatitis, HBV associated
55
M
Excluded (CCC)
NA
Normal
39
F
CRLM
MD
Normal
63
M
CRLM
Mucinous adenocarcinoma
Normal
36
M
HCC
MD
Chronic hepatitis, HBV-associated
69
M
CRLM
MD
Normal
76
F
Excluded (CCC)
NA
Macronodular cirrhosis, HBVassociated
53
M
CRLM
MD
Normal
33
F
CRLM
MD
Normal
65
M
Excluded (poor quality spectra)
MD
Normal
57
M
HCC
MD
Macronodular cirrhosis, HBVassociated
49
M
CRLM
MD
Normal
68
F
CRLM
MD
Normal
74
M
CRLM
MD
Normal
51
M
HCC
MD
Micronodular cirrhosis, etiology uncertain
68
M
CRLM
MD
Normal
60
M
CRLM
PD
Normal
72
M
HCC
MD
Steatosis
54
F
HCC
MD
Chronic hepatitis, HBV associated
35
M
Excluded (combined)
MD
Chronic hepatitis, HBV-associated
58
M
CRLM
MD
Normal
52
M
HCC
MD
Macronodular cirrhosis, HBVassociated
58
M
HCC
MD
Chronic hepatitis, HBV-associated
59
F
HCC
MD
Chronic hepatitis, HBV-associated
M male, F female, CRLM colorectal liver metastases, HCC hepatocellular carcinoma, combined combined hepatocellular-cholangiocarcinoma, CCC cholangiocarcinoma, NA not applicable, PD poorly differentiated, MD moderately differentiated, HBV hepatitis virus B
3.2 HRMAS 1H NMR Spectra and Statistical Analysis The stacked HRMAS 1H NMR spectra of non-tumorous hepatic parenchyma, HCC, and CRLM are shown in Fig. 1. Visual inspection of the spectra showed clear differences in metabolites profiles between HCC and CRLM as well as between each tumor and the non-tumorous hepatic parenchyma.
123
26
S. Y. Kim et al.
Fig. 1 Stacked HRMAS 1H NMR spectra of HCC, CRLM, and non-tumorous hepatic parenchyma. a Non-tumorous hepatic parenchyma, b hepatocellular carcinoma (HCC), and c colorectal liver metastases (CRLM)
In HCC, CRLM, and non-tumorous hepatic parenchyma, the OPLS-DA score plot showed a good splitting with a good description of the data (R2X = 0.929, R2Y = 0.765) and a good prediction accuracy for new data (Q2 = 0.508) (Fig. 2). Spectra from HCC lay between non-tumorous hepatic parenchyma and CRLM. HCC produced a different metabolic profile from that of CRLM. In OPLS-DA analysis, the two components models provided an excellent description of the data (R2X = 0.922, R2Y = 0.927) and a good prediction accuracy for new data (Q2 = 0.569). This was demonstrated by an excellent separation of these two groups in the OPLS-DA score plots (Fig. 3a). The loading plots demonstrated the marker metabolites that were responsible for the separation of HCC and CRLM (Fig. 3b). The mean metabolite quantities of marker metabolites responsible for separation of HCC and CRLM are shown in Table 3. Compared with CRLM, HCC showed significantly elevated levels of glucose (P \ 0.01), and GPC (P \ 0.01), and decreased levels of hypoxanthine (P = 0.04). When we compared HCC and non-tumorous hepatic parenchyma, the OPLS-DA score plots discriminated HCC from non-tumorous hepatic parenchyma with a good description of the data (R2X = 0.913, R2Y = 0.857) and a low prediction accuracy for new data (Q2 = 0.24) (Fig. 4a). Among the marker metabolites responsible for the separation of HCC and non-tumorous hepatic parenchyma (Fig. 4b), the levels of alanine and PC were significantly elevated in HCC (Table 3). The level of fatty
123
Differentiation Between Hepatocellular
27
Fig. 2 OPLS-DA score plot to compare metabolites in HCC, CRLM, and non-tumorous hepatic parenchyma The OPLS-DA score plot shows a good splitting of the spectra from HCC, CRLM, and the non-tumorous hepatic parenchyma. Note that the spectra from HCC are located between non-tumorous hepatic parenchyma and CRLM. (Filled square) non-tumorous hepatic parenchyma, (filled diamond) hepatocellular carcinoma (HCC), (filled circle) colorectal liver metastases (CRLM)
acid was decreased in HCC, even though we did not compare the exact level as fatty acid consists of several compounds. Regarding the discrimination of CRLM from non-tumorous hepatic parenchyma, OPLS-DA analysis provided an excellent description of the data (R2X = 0.91, R2Y = 0.95) and a good prediction accuracy for new data (Q2 = 0.842). The OPLSDA score plots demonstrated a clear separation of these two groups (Fig. 5a). Based on the loading plots (Fig. 5b) and comparison of the mean relative concentrations of the marker metabolites, CRLM showed a significant elevation in the levels of alanine, glutamate, hypoxanthine, lactate, leucine, and valine, while with a significant decrease in the levels of glucose and GPC (Table 3). The level of fatty acid was decreased in CRLM, although we did not compare the exact level as fatty acid consists of several analogue compounds.
4 Discussion Our study demonstrated that HRMAS 1H NMR spectroscopy was able to discriminate HCC, CRLM, and non-tumorous hepatic parenchyma. Interestingly, on OPLS-DA score plots comparing the two tumors and non-tumorous hepatic parenchyma, HCC was located between CRLM and non-tumorous hepatic parenchyma. This can be explained by the fact that HCC bears a resemblance to
123
28
S. Y. Kim et al.
Fig. 3 OPLS-DA score plots and the loading plots used to compare metabolites in HCC and CRLM. a The OPLS-DA plot demonstrates the good separation of the metabolites from HCC and CRLM. b The loading plot identifies the marker metabolites that are responsible for the separation of CRLM from HCC. (Filled diamond) hepatocellular carcinoma (HCC), (filled circle) colorectal liver metastases (CRLM)
123
7.061 0.569
1.097 0.159
2.244 0.224
1.429 0.406
0.573 0.138
2.331 0.502
0.734 0.196
0.663 0.151
1.422 0.396
3.170 0.562
Methionine
O-Phosphocholine
Phenylalanine
Tyrosine
Valine
sn-Glycero-3-phosphocholine
3.821
1.533
0.486
0.663
0.968
0.554
1.202
0.334
0.137
0.044
0.086
0.081
0.051
0.128
0.177
1.290
0.094
0.020
0.476
0.567
0.124
0.305
2.140
0.336
0.320
0.725
0.182
SE
0.096
* Statistically significant
NA
0.001*
[0.999
0.022* \0.001*
[0.999 [0.999
0.198
[0.999
\0.001*
NA
\0.001*
NA
NA 0.894
NA
0.380
NA
NA
0.111
NA
NA
0.252
NA
0.018* 0.012*
0.116 NA
0.117
[0.999
0.303
0.039* [0.99
NA
NA
NA
\0.001*
\0.001*
0.251
\0.001*
NA
NA
0.185 [0.999
NA
0.090
0.963
0.249
[0.999
\0.001* NA
0.049* 0.240
NA [0.999
NA [0.999
0.198 [0.999
HCC vs. CRLM HCC vs. nontumorous CRLM vs. nontumorous hepatic parenchyma hepatic parenchyma
HCC hepatocellular carcinoma, CRLM colorectal liver metastases, SE standard error, NA not available comparison
0.788 0.146
1.141 0.222
1.191 0.254
0.740 0.079
1.829 0.238
2.312
Lysine
3.310 0.295
2.279 0.543
Leucine
1.081
0.978 0.246
1.446 0.151
6.600 0.269
20.821 2.228 22.648 2.563 15.138
0.869 0.157
5.362
2.505
4.413
2.373
3.626
Isoleucine
Hypoxanthine
3.449 0.523
1.826 0.304
7.648 0.501
6.024 0.603
Lactate
4.790 0.813
0.311 0.040
Glycine
2.878 0.294
5.945 1.218
4.469 0.958
23.575 3.660
Glucose
Glutamate
Glycerol
6.650 1.314 26.271
1.216 0.344
Glutamine
1.510 0.113
5.426 0.768
Choline
1.316 3.977
Alanine
2.244 0.581
3.101 0.753
0.772 0.134
3.412 1.526
Mean
Adipate
SE
Nontumorous hepatic parenchyma P values
3-Hydroxybutyrate
Mean
Mean
SE
CRLM
HCC
Table 3 Relative concentrations for marker metabolites which contributed to the classification of HCC, CRLM, and nontumorous hepatic parenchyma
Differentiation Between Hepatocellular 29
123
30
S. Y. Kim et al.
Fig. 4 OPLS-DA score plot to compare metabolites in HCC and non-tumorous hepatic parenchyma. a The OPLS-DA score plot clearly discriminates HCC from non-tumorous hepatic parenchyma. b The loading plot identifies the marker metabolites that are responsible for the separation of HCC from the nontumorous hepatic parenchyma. (Filled diamond) hepatocellular carcinoma (HCC), (filled square) nontumorous hepatic parenchyma
123
Differentiation Between Hepatocellular
31
Fig. 5 OPLS-DA score plot to compare metabolites in CRLM and non-tumorous hepatic parenchyma. a The OPLS-DA score plot demonstrates the clear separation of CRLM and the non-tumorous hepatic parenchyma. b The loading plot identifies the marker metabolites that are responsible for the separation of CRLM from the non-tumorous hepatic parenchyma. (Filled circle) colorectal liver metastases (CRLM), (filled square) non-tumorous hepatic parenchyma
123
32
S. Y. Kim et al.
hepatocyte in its morphology, although CRLM does not. As demonstrated in the HRMAS 1H NMR spectroscopic analysis of brain tumors [29], tumors with related histology have similar metabolite profiles which reflect the known biological and morphological behavior of tumors. HRMAS 1H NMR spectroscopy also identified the differences in the metabolite profiles of HCC and CRLM, which are the most common hepatic tumors. HCC showed significantly elevated levels of glucose, and GPC, and decreased levels of hypoxanthine, compared with CRLM. We speculated that the differences in the protein patterns in HCC and CRLM demonstrated in a previous study [13] was reflected in the metabolomic differences in our study. Even though the discriminating metabolites between each type of tumor and nontumorous hepatic parenchyma differed, the increase of lactate was commonly found in both types of tumor. Lactate, a widely described metabolite of anaerobic glycolysis that increases rapidly during hypoxia and ischemia, was frequently reported to increase in a range of cancer cells [3] including the results of HRMAS 1 H NMR spectroscopic analysis of HCC [10] and primary colorectal cancers [28]. In addition, the level of fatty acid decreased in both types of tumor and was thus consistent with what was seen in previous HCC [10] and primary colorectal cancer studies [28]. Decrease in the level of fatty acid was likely to be associated with a higher metabolic turnover as well as the requirement in membrane biosynthesis for cell propagation leading to higher lipid utilization [10, 28]. Previous studies demonstrated that ex vivo HRMAS 1H NMR spectra had the potential to be translated to in vivo magnetic resonance spectra obtained by the same physical principle, although at a much lower resolution [30–32]. However, the trend in changes of the choline-containing compound in both HCC and CRLM according to the ex vivo spectra in our study differed from that of in vivo magnetic resonance spectroscopy [14, 33, 34] which demonstrated an elevation in choline levels. The choline compound is generally regarded as the marker of increased membrane turnover and is expected to increase in malignant tumors [3]. In our study, HRMAS 1H NMR spectroscopy showed that the choline and GPC levels were decreased, whereas the PC levels were elevated in both HCC and CRLM, compared with that seen in non-tumorous hepatic parenchyma. Due to limitations in resolution, in vivo spectroscopy of tumors cannot determine individual cholinecontaining compounds including choline, GPC and PC, but provide the sum of all choline-containing compounds. In contrast, given the combination of high spectral resolution and the possibility to analyze intact tissue specimens, ex vivo HRMAS 1 H NMR spectroscopy allows determination of each choline-containing compound in detail. Although many in vivo magnetic resonance spectra studies showed an elevated level of total choline compounds in malignancy [14, 33, 34], HRMAS 1H NMR spectroscopy studies showed more complex results [35]. Some study results demonstrated a decrease in the level of GPC and an increased level of PC in malignant lesions, which is consistent with the results of our study [10, 36]. The poorer correlation between in vivo and ex vivo magnetic resonance spectra was more frequently found on smaller peaks compared with the larger peaks [31]. However, as changes in a level of choline compounds in previous studies were
123
Differentiation Between Hepatocellular
33
inconsistent and diverse, further studies would be warranted to confirm the changes in individual choline-containing compounds in malignancy [35]. In our study, a significant elevation of alanine in HCC compared with that in nontumorous hepatic parenchyma was consistent with the research regarding HRMAS 1 H NMR spectroscopic analysis of HCC [10]. The elevation of alanine was possibly due to enhanced glycolysis and the effect of the TCA pathway [3]. According to the previous studies, there are qualitative and quantitative changes in the genes and proteins in metastasizing colorectal cancer compared with those seen in primary colorectal cancer [37–39]. Understanding those changes is of great importance for the early detection and treatment of metastasis. To our knowledge, the metabolomic profiles of CRLM have not previously been reported. Even though we did not compare primary colorectal cancer and CRLM, the results from our study could be a base for future studies attempting to identify the differences in the metabolite profiles of primary and metastatic colorectal cancer. The metabolite profiles demonstrated in our study have the potential to be clinically applied in the future. Since in vivo spectroscopy is obtained by the same physical principle with HRMAS 1H NMR spectroscopy, translational research is feasible from excised tissue to non-invasive examinations in humans [30–32]. The progression of high magnetic field in vivo MR scanner for the human could allow high-resolution spectroscopic studies in the future. However, current in vivo magnetic resonance spectroscopy of human is not obtainable at the same spectral resolution to ex vivo spectroscopy. The metabolic profiles of most common hepatic tumors demonstrated in our study with HRMAS 1H NMR may help to identify peculiar changes crowded regions of the magnetic resonance spectrum. Interpretation of the significance of specific metabolite within the context of treatment decision or the prediction of prognosis of the patient would be the future direction for our study, even though the current study is limited due to the small number of the patients. Our study has limitations. First, it should be noted that our study was limited in its sample size. The significance of our statistical analysis must be considered in the particular context of the preliminary study in which the number of cases presented was still relatively small. Considering the many factors affecting the metabolite changes in the tumor such as the histologic grade and heterogeneity of the tumor microenvironment, further study with a large number of samples will be needed to establish the metabolite panel of HCC and CRLM. Second, even though the different metabolite profiles separating HCC and CRLM were indentified, we were not able to determine the meaning of all of the metabolites. For a detailed characterization and clarification of the significance of the metabolite differences in both tumors, further study combined with known prognostic factors, including genetic and proteomic information will be needed. Third, we did not perform absolute quantification of the metabolites, because absolute metabolite concentration quantification in HRMAS 1H NMR spectroscopy is a very challenging task and is associated with several sources of error [23, 24]. Due to these reasons, relative quantification normalized by the sum of all integral regions is generally preferred in HRMAS 1H NMR spectroscopy [23, 24]. However, it would be helpful to find an invariant metabolite in the liver which could then be used as a standard metabolite
123
34
S. Y. Kim et al.
such as creatine in brain magnetic resonance spectroscopy. Fourth, instead of analyzing fresh samples, we preserved them using snap-freezing method, which could affect the metabolite profiles. However, a previous report demonstrated that few changes in metabolite signals were observed in HRMAS 1H NMR spectra between the fresh liver specimens and snap-frozen liver specimens [18]. In conclusion, HRMAS 1H NMR spectroscopy successfully differentiated HCC from CRLM. In addition, HRMAS 1H NMR spectroscopy can also metabolically characterize HCC and CRLM, which could be used for a future study regarding metabolite profiling of both tumors. Acknowledgments This study was supported by grant no. 11-2009-025 from the Seoul National University Bundang Hospital Research Fund, grant no. A091075 of the Korea Health Technology R&D Project, Ministry of Health & Welfare, Korea, grant no. 2008-03876 of the Nuclear R&D program of the Korea Science and Engineering Foundation funded by the Ministry of Education, Science and Technology of Korea, grant no. HI10C2014 from the Korean Health Technology R&D Project, Ministry for Health and Welfare, Republic of Korea, grant no. 09-384 from the ASAN Institute for Life Science, Seoul, Korea, and grant no. 2012R1A1A1012731 by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology.
References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20.
J.L. Spratlin, N.J. Serkova, S.G. Eckhardt, Clin. Cancer Res. 15, 431–440 (2009) O. Fiehn, Plant Mol. Biol. 48, 155–171 (2002) J.L. Griffin, J.P. Shockcor, Nat. Rev. Cancer 4, 551–561 (2004) W.B. Dunn, N.J. Bailey, H.E. Johnson, Analyst. 130, 606–625 (2005) M.E. Bollard, S. Garrod, E. Holmes, J.C. Lindon, E. Humpfer, M. Spraul, J.K. Nicholson, Magn. Reson. Med. 44, 201–207 (2000) N.J. Waters, E. Holmes, C.J. Waterfield, R.D. Farrant, J.K. Nicholson, Biochem. Pharmacol. 64, 67–77 (2002) O.M. Rooney, J. Troke, J.K. Nicholson, J.L. Griffin, Magn. Reson. Med. 50, 925–930 (2003) I.F. Duarte, E.G. Stanley, E. Holmes, J.C. Lindon, A.M. Gil, H. Tang, R. Ferdinand, C.G. McKee, J.K. Nicholson, H. Vilca-Melendez, N. Heaton, G.M. Murphy, Anal. Chem. 77, 5570–5578 (2005) B. Martinez-Granados, D. Monleon, M.C. Martinez-Bisbal, J.M. Rodrigo, J. del Olmo, P. Lluch, A. Ferrandez, L. Marti-Bonmati, B. Celda, NMR Biomed. 19, 90–100 (2006) Y. Yang, C. Li, X. Nie, X. Feng, W. Chen, Y. Yue, H. Tang, F. Deng, J. Proteome Res. 6, 2605–2614 (2007) H. Gao, Q. Lu, X. Liu, H. Cong, L. Zhao, H. Wang, D. Lin, Cancer Sci. 100, 782–785 (2009) P. R. Ros, S. M. Erturk, in Textbook of Gastrointestinal Radiology, ed. by R. M. Gore, M. S. Levine 3rd edn. (Saunders, Philadelphia, 2008), p. 1623–1662 C. Melle, G. Ernst, B. Schimmel, A. Bleul, F. von Eggeling, PLoS One 3, e3767 (2008) C.Y. Chen, C.W. Li, Y.T. Kuo, T.S. Jaw, D.K. Wu, J.C. Jao, J.S. Hsu, G.C. Liu, Radiology. 239, 448–456 (2006) W. Chen, Y. Zu, Q. Huang, F. Chen, G. Wang, W. Lan, C. Bai, S. Lu, Y. Yue, F. Deng, Magn. Reson. Med. 66, 1531–1540 (2011) L. Eriksson, E. Johansson, N. Kettaneh-Wold, J. Trygg, C. Wikstrom, S. Wold. in Multi- and megavariate data analysis (Umetrics AB, Umea˚, 2006) J. Trygg, S.J. Wold, Chemometric. 16, 119–128 (2002) N.J. Waters, S. Garrod, R.D. Farrant, J.N. Haselden, S.C. Connor, J. Connelly, J.C. Lindon, E. Holmes, J.K. Nicholson, Anal. Biochem. 282, 16–23 (2000) S. Meiboom, D. Gill, Rev. Sci. Instrum. 20, 688–691 (1958) P. Mazzei, A. Piccolo, L. Nugnes, M. Mascolo, G. De Rosa, S. Staibano, NMR Biomed. 23, 1137–1145 (2010)
123
Differentiation Between Hepatocellular
35
21. M.N. Triba, A. Starzec, N. Bouchemal, E. Guenin, G.Y. Perret, L. Le Moyec, NMR Biomed. 23, 1009–1016 (2010) 22. H. Wu, X. Zhang, X. Li, Y. Wu, F. Pei, Anal. Biochem. 339, 242–248 (2005) 23. Y. Wang, M.E. Bollard, H. Keun, H. Antti, O. Beckonert, T.M. Ebbels, J.C. Lindon, E. Holmes, H. Tang, J.K. Nicholson, Anal. Biochem. 323, 26–32 (2003) 24. Y. Wang, M.E. Bollard, J.K. Nicholson, E. Holmes, J. Pharm. Biomed. Anal. 40, 375–381 (2006) 25. N. D. Theise, Y. N. Park, M. P. Curado, M. Sakamoto, S. Franceschi, M. Torbenson, P. Hytiroglou, A. Wee, M. Kudo, in WHO Classification of Tumours of the Digestive System, ed. by F. T. Bosman, F. Carneiro, R. H. Hruban (IARC, Lyon, 2010), p. 205–216 26. E. Westman, L.O. Wahlund, C. Foy, M. Poppe, A. Cooper, D. Murphy, C. Spenger, S. Lovestone, A. Simmons, J. Alzheimers Dis. 22, 171–181 (2010) 27. A. Imperiale, K. Elbayed, F.M. Moussallieh, N. Reix, M. Piotto, J.P. Bellocq, B. Goichot, P. Bachellier, I.J. Namer, Endocr. Relat. Cancer 20, 705–716 (2013) 28. E.C. Chan, P.K. Koh, M. Mal, P.Y. Cheah, K.W. Eu, A. Backshall, R. Cavill, J.K. Nicholson, H.C. Keun, J. Proteome Res. 8, 352–361 (2009) 29. M. Wilson, N.P. Davies, M.A. Brundler, C. McConville, R.G. Grundy, A.C. Peet, Mol. Cancer 8, 6 (2009) 30. C. Boesch, Clin. Sci. (Lond.) 116, 401–402 (2009) 31. J.F. Cobbold, J.H. Patel, R.D. Goldin, B.V. North, M.M. Crossey, J. Fitzpatrick, M. Wylezinska, H.C. Thomas, I.J. Cox, S.D. Taylor-Robinson, J. Hepatol. 52, 16–24 (2010) 32. J.F. Cobbold, Q.M. Anstee, R.D. Goldin, H.R. Williams, H.C. Matthews, B.V. North, N. Absalom, H.C. Thomas, M.R. Thursz, R.D. Cox, S.D. Taylor-Robinson, I.J. Cox, Clin. Sci. (Lond.) 116, 403–413 (2009) 33. H. Xu, X. Li, Z.H. Yang, J.X. Xie, Acad. Radiol. 13, 1532–1537 (2006) 34. C.W. Li, Y.C. Kuo, C.Y. Chen, Y.T. Kuo, Y.Y. Chiu, F.O. She, G.C. Liu, Magn. Reson. Med. 53, 770–776 (2005) 35. S. Moestue, B. Sitter, T.F. Bathen, M.B. Tessem, I.S. Gribbestad, Curr. Top. Med. Chem. 11, 2–26 (2011) 36. B. Sitter, S. Lundgren, T.F. Bathen, J. Halgunset, H.E. Fjosne, I.S. Gribbestad, NMR Biomed. 19, 30–40 (2006) 37. B. Yu, S.Y. Li, P. An, Y.N. Zhang, Z.J. Liang, S.J. Yuan, H.Y. Cai, World J. Gastroenterol. 10, 2652–2656 (2004) 38. S.D. Finkelstein, R. Sayegh, S. Christensen, P.A. Swalsky, Cancer 71, 3827–3838 (1993) 39. M. Conzelmann, U. Linnemann, M.R. Berger, Int. J. Colorectal Dis. 20, 137–146 (2005)
123
