Metabolomics DOI 10.1007/s11306-013-0601-2

ORIGINAL ARTICLE

High resolution-magic angle spinning (HR-MAS) NMR-based metabolomic fingerprinting of early and recurrent hepatocellular carcinoma Antonio Solinas • Matilde Chessa • Nicola Culeddu • Maria Cristina Porcu • Giuseppe Virgilio • Francesco Arcadu • Angelo Deplano • Sergio Cossu • Domenico Scanu • Vincenzo Migaleddu

Received: 10 June 2013 / Accepted: 18 October 2013 Ó Springer Science+Business Media New York 2013

Abstract In order to elucidate the metabolite changes associated with hepatocellular carcinoma (HCC) oncogenesis and progression, we compared the profiles obtained by 1D proton HRMAS NMR spectroscopy of 51 needle biopsies (14 primary nodules, 14 recurrent, and 23 paired cirrhotic specimens). The diagnosis of HCC was based on 2 concordant imaging techniques and was confirmed by histology in 20 cases. Spectroscopy was performed using a Bruker AVANCE II 600 spectrometer. One-dimensional proton spectra were acquired using water-suppressed (noesygppr) pulse and spin-echo CPMG sequences. Signals were assigned by BBIOREFCODE and were confirmed by HSQC. Statistics was based on the SIMCA P package. Orthogonal projection to latent structure (OPLS-DA) showed a clear separation between tumor and cirrhosis.

Electronic supplementary material The online version of this article (doi:10.1007/s11306-013-0601-2) contains supplementary material, which is available to authorized users.

This difference was maintained when the analysis of paired samples from primary to recurrent nodules was split. OPLS-DA of primary and recurrent nodules also showed a significant difference. The relationship between metabolite profile and HCC volume was evaluated comparing the spectra obtained in tumors B2 cm (n = 15) and in those larger than 2 cm (n = 11). Univariate comparison of the most relevant metabolites showed that: (1) increased choline, TMAO, and decreased saturated fatty acids differentiate HCC from the surrounding tissue; (2) increased lactate and myo-inositol differentiate recurrent from primary HCC; (3) decreased saturated fatty acids characterize large HCC nodules. Keywords Early hepatocellular carcinoma
Recurrent hepatocellular carcinoma
1D HRMASNMR
Ex-vivo spectroscopy
Metabolomic fingerprinting

A. Solinas Department of Biomedical Sciences, University of Sassari School of Medicine, Viale San Pietro 12, 07100 Sassari, Italy

A. Deplano Unit of Internal Medicine, Department of Medicine, Community Hospital of Lanusei, Lanusei, Italy

M. Chessa
N. Culeddu (&)
M. C. Porcu Institute of Biomolecular Chemistry – National Council of Research – Italy, CNR – National Research Council of Italy, Traversa la Crucca 3, 07040 Sassari, Italy e-mail: [email protected]

S. Cossu Unit of Morbid Anatomy, Department of Pathology, Community Hospital of Nuoro, Via Mannironi, Nuoro, Italy

G. Virgilio
V. Migaleddu Sardinian Mediterranean Imaging Research Group – (SMIRG) Foundation, Via Stintino 2, 07100 Sassari, Italy

D. Scanu Unit of Interventional Echography, Department of Medicine, Community Hospital of Nuoro, Macomer, Nuoro, Italy

F. Arcadu Unit of Internal Medicine and Gastroenterology, Department of Medicine, Community Hospital of Nuoro, Via Mannironi, Nuoro, Italy

123

A. Solinas et al.

1 Introduction Surveillance of patients at risk of hepatocellular carcinoma (HCC) has resulted in early treatment and in improved survival (Sangiovanni et al. 2004; Zhang et al. 2004; Altekruse et al. 2012; Forner et al. 2012; Livraghi et al. 2008). Currently, the expected 5 year survival rate following liver resection for primary HCC B2 cm, in patients with ChildPugh A cirrhosis, ranges between 80 and 90 %. Long term outcome of these patients depends uniquely on the development of new tumoral lesions. Among the factors relevant to recurrence, the volume of the primary tumour is critical (Chun et al. 2011), as microvascular invasion was observed in 27 % of nodules B2 cm, and in 80 % of those [3 cm (Minquez et al. 2011; Roayaie et al. 2013). Recurrent nodules are usually more aggressive than primary small HCC, in terms of vascular invasion, growth rate, and multinodular pattern (Chun et al. 2011; Park et al. 2008). The progression of HCC from single slow-growing nodule to aggressive metastatic tumor is explained by the occurrence of cumulative genomic alterations (Negrini et al. 2010; Calvisi et al. 2004). However, the corresponding metabolic changes associated with HCC oncogenesis and progression are unclear. Metabolomics encompasses identification and quantification of low (\1 kDa) molecular weight metabolites in a biological system. It is known that the metabolism of cancer cells differs from that of normal cells in several aspects, including glycolysis (Gogvadze et al. 2009; Lo´pez-La´zaro 2008) and synthesis and catabolism of fatty acids (Griffin and Shockcor 2004; Ng et al. 2011). A few studies have addressed HCC metabolomics, comparing ‘‘ex vivo’’ metabolite profiles of human HCC with corresponding adjacent tissue. These studies differ in regard to their analytical strategy and aims. Yang et al. (2007), using NMR spectroscopy, provided the first pattern recognition analysis differentiating HCC from non-tumoral tissue, and high grade HCC from low grade HCC. These findings were confirmed by Paris et al. (2010). Beyoglu et al. (2013), on the other hand, provided the first target analysis of HCC using gas chromatography coupled to mass spectrometry (GCMS). This study shows that HCC undergoes metabolic remodeling including a four-fold increase of aerobic glycolysis over mitochondrial oxidative phosphorylation. This finding is the expression of increased energy demand of neoplastic cells in comparison to the surrounding tissue. In addition, distinct HCC subgroups, classified according to transcriptomics, are characterized by different tissue concentrations of monoacylglycerols. Other studies, focused on lipid metabolism, found that increased lipogenesis, induced by AKT-mTORC1-RPS6 signaling promotes the development of HCC (Calvisi et al. 2011) and lipid metabolites of stearoyl-CoA desaturase are associated with

123

Table 1 Baseline characteristics of 25 patients with HCC Primary HCC

Recurrent HCC

No of patients

12

13

Age (median and range)

61 (52–68)

72 (64–80)

Gender m/f

12/0

13/0

Child-pugh A5/B7

11/1

No of patients with HCV infection

6

8

No of patients with HBV infection

2

2

No of patients with alcohol abuse Steatohepatitis

4 0

2 1

AST (IU/mL) (median and range)

63 (29–156)

102 (25–215)

ALT (IU/mL) (median and range)

76 (17–171)

93 (16–146)

PT/INR (median and range)

1.2 (1.1–1.4)

1.1 (1.1–1.3)

Bilirubin (mg/dL) (median and range)

0.8 (0.4–2.7)

1.2 (0.6–1.8)

Platelets (109/L) (median and range)

164 (100–248)

121 (69–233)

Albumin (g/dL) (median and range)

3.8 (3–4.1)

3.8 (2.84–4.6)

Creatinine (mg/dL) (median and range)

0.7 (0.5–0.9)

0.7 (0.5–1)

Fasting blood sugar (mg/dL) (median and range)

101 (79–195)

94 (72–174)

Diabetes

4/12

4/13

Alpha-fetoprotein (ng/mL) (median and range)

13.4 (0.8–27.4)

39.7 (2.2–3,159)

No of patients with a solitary nodule

9/12

7/13

No of patients with 2/3 synchronous nodules

3/0

3/3

Maximum size of nodules (mm) (median and range)

20 (14–45)

17 (10–35)

aberrant palmitate signaling in aggressive HCC samples (Budhu et al. 2013). All these studies were based on large HCC nodules. In fact, the mean level of alpha-fetoprotein in the study of Yang et al. (2007) was 6,620 ng/mL, the mean size of the tumor in the study of Beyoglu et al. (2013) was 7.1 cm and, in the study of Budhu, 93 % of the patients had nodules greater than 3 cm in diameter. In addition, all specimens were from resected livers. High resolution magic angle spinning NMR (HR-MAS NMR) is a powerful analytical technique, which requires minimal sample quantity and virtually no preparation. MartinezGranados et al. (2006) have shown that HR-MAS NMR spectroscopy identifies the metabolite profiles of needle biopsies of human liver. This finding opened up to applications of metabolomics in a standard clinical setting. It was our aim to evaluate whether metabolomics, based on HR-MAS NMR spectroscopy of needle biopsies is a

HR-MAS NMR-based metabolomic fingerprinting

suitable tool in the diagnosis of early HCC and in the evaluation of its progression. This study, consisting of a cross sectional analysis of ex vivo tumoral and surrounding tissue of primary and recurrent nodules of distinct size, represents a proof of principle of HR-MAS NMR spectroscopy in diagnosis of small HCC.

2 Experimental section 2.1 Sample collection This study includes 25 patients, divided into 2 groups. The first group consisted of 12 consecutive patients with compensated cirrhosis and no previous history of HCC, and with a newly detected nodule by standard ultrasound surveillance. In all these patients, the subsequent contrast enhanced ultrasound (CEUS) and multidetector computer aided tomography (CT) scans showed arterial and delayed phase typical of HCC. In 3 of them a second synchronous lesion was detected. The second group consisted of 13 patients with compensated cirrhosis and HCC who had received surgical resection or radiofrequency ablation and in whom, after a period of 3–24 months free of recurrence, a new nodule, or more than a single nodule (2 in 3 patients and 3 in 3 patients) were detected. In this group also, the diagnosis of recurrent HCC was confirmed by CEUS and contrast enhancement CT scan. The baseline characteristics of the 2 groups are reported in Table 1. After radiological diagnosis, all patients underwent radiofrequency ablation. In this setting, under general anaesthesia, ultrasound guided 18 gauge core needle biopsies with 22 mm penetration (BARD Monopty UK) were performed. Paired biopsies of the extra nodular parenchyma were performed in 23 cases. Split specimens obtained by needle biopsies were immediately quenched and stored in liquid nitrogen until spectroscopy was performed. The second sample was processed for histology. All procedures were approved by the Ethical Committee of the Azienda Sanitaria no.3 Nuoro-Italy. This study was performed in accordance with the Declaration of Helsinki and was formally approved by the regulatory authorities of our institution. All patients gave their consent to the procedure. 2.2 High resolution magic angle spinning (HR-MAS) NMR spectroscopy High-resolution proton NMR experiments were performed at 277 K using a Bruker AVANCE II 600 spectrometer. The biopsy material was placed into the HRMAS rotor and was added with 5 lL of phosphate buffered saline (PBS), pH 7.2, and 10 % D2O and 3-(Trimethylsilyl)propionic-2,2,3,3-d4 acid sodium salt (TSP, 0.01 %) for referencing purposes. Samples were spun at 4,000 Hz. One dimensional (1D) proton spectra were acquired using a water-suppressed (noesygppr)

pulse sequence with 2.5 s water-presaturation during relaxation delay, 8 kHz spectral width, 32 k data points, 32–64 scans. A water-suppressed spin-echo Carr-Purcell-MeiboomGill sequence (CPMG-pr) was obtained using 2.5 s water presaturation during relaxation delay, 1 ms echo time (s), and 360 ms total spin–spin relaxation delay (2ns), 8 kHz spectral width, 32 k data points, 128 scans. 2D 1H- 13C-HSQC spectra were recorded on selected samples to facilitate spectral assignment, with a spectral width of 9,920 Hz in the F2 dimension and 33,936 Hz in the F1 dimension, a data matrix of 1,400 9 512 data points and 48 scans per increment. Data were processed using the Bruker TOPSPIN 3.2. Metabolite identification was performed using BBIOREFCODE 2.0.2 database (Bruker Bio-Spin GmbH Rheinstetten, Germany), HMDB database (Wishart and Jewison 2013) and literature references (Martinez-Granados et al. 2006). Assignment of most significant metabolites was confirmed by 2D TOCSY and 2D 1H- 13C-HSQC spectra. 2.3 Data analysis Prior to chemometric analysis, data were normalized by setting the region between 9–0.54 ppm of each spectrum to 100. Each spectrum was segmented in 224 consecutively integrated regions 0.04 ppm wide using Analysis of MIXtures (AMIX) (Bruker GmbH, Karlsruhe, Germany). The generated ASCII file was imported into Microsoft EXCEL for the addition of labels. The bucket matrix was imported into SIMCA-P software version 13.0, (Umetrics AB, Umea˚, Sweden) for statistical analysis. All data were mean-centered. Supervised analysis was performed applying orthogonal partial least square discriminant analysis (OPLS-DA), which implies a rotation of the corresponding PLS-DA models and simplifies them concentrating the information in one predictive component maintaining the same predictive ability (Trygg and Wold 2002; Bylesjo¨ et al. 2006). The number of the orthogonal components in the models was determined by the ‘‘autofit’’ routine of SIMCA-P. In order to avoid model overfitting, the corresponding PLSDA models were validated by 300 times permutation tests. The prediction strength of the model was evaluated by leave one out analysis. Univariate comparison of peak integrals related to distinct metabolites was performed using non parametric U Mann–Whitney analysis.

3 Results and discussion 3.1 Histological diagnosis of hepatocellular carcinoma 15 biopsies were obtained from 12 patients with primary focal lesions. In these cases, histology showed the features of HCC in 10 out of 15 samples and was inconclusive in

123

A. Solinas et al. Fig. 1 a 1H HR-MAS NMR spectra of biopsies from parenchyma and nodule, b HSQC spectra showing assignments of choline and its derivatives

the remaining 5. Edmondson Grade was 1 in 2 cases, 2 in 6 cases, and 3 in 2 cases. 15 biopsies were obtained from 13 patients with recurrent nodules. The histological features of HCC were observed in 10 biopsies. Edmondson Grade was 1 in 2 cases, 2 in 6 cases, and 3 in 2 cases. The histological diagnosis was inconclusive in the remaining 5 cases. 25 biopsies from the corresponding parenchyma showed the features of cirrhosis in all cases. The biopsies taken from the surrounding cirrhotic tissue of the 14 HCV and 4 HBV related cirrhosis did not show significant differences in terms of fibrous tissue deposition, lymphocytic infiltrate and hepatocellular changes. By contrast, in 3 of the remainders related to alcohol abuse (6 cases)

123

and steatohepatitis (1 case), macrovescicular steatosis was observed. 3.2 NMR spectra and metabolomics of neoplastic nodules and adjacent parenchyma A total of 53 biopsies were obtained from 25 patients: 30 were from nodules (5 from patients with more than one lesion) and 25 were from the surrounding tissue. Four specimens (2 nodules and parenchyma samples) did not yield reliable spectra. The five specimens from patients with more than one lesion (3 primary and 2 recurrent nodules) were all included in the analysis, in the assumption we were dealing

HR-MAS NMR-based metabolomic fingerprinting Fig. 2 Scores plot and S-plot resulting from OPLS-DA applied to the 1H HR-MAS NMR spectra of biopsies from parenchyma (purple diamonds) and nodule (turquoise pentagon). In S-plot red circles refer to metabolites in Table 2 (Color figure online)

with distinct lesions, rather than with repeated biopsies of the same lesion. Therefore, 51 tissue samples 28 from the neoplastic lesions and 23 from the parenchyma were available for analysis by HRMAS-NMR spectroscopy. Figure 1 shows representative spectra obtained from a neoplastic nodule and the corresponding adjacent tissue and highlights partial

metabolite assignment in the chemical shift region ranging from 0.8 to 4.68 ppm. In order to evaluate whether these differences were also statistically significant, we built a pattern recognition model, comparing the spectra obtained from the tumoral nodules and the corresponding adjacent areas.

123

A. Solinas et al.

The paired spectra were evaluated by the orthogonal projections to latent structures-discriminant analysis (OPLSDA) (Trygg and Wold 2002; Bylesjo¨ et al. 2006) with one predictive and four orthogonal components. By means of this analysis, a statistically significant separation between tumor and parenchyma derived spectra was obtained. Figure 2 shows the corresponding 2D score plot of the predictive component versus the first orthogonal component. As shown in the figure, HCC nodules tended to cluster into 2 subgroups, whereas liver cirrhotic tissue was more dispersed with 2 overlapping samples and 1 outlier. It is of note, that specimens from the 5 nodules with multiple lesions yield to distinct spectra, suggesting a different profile. In addition differences in liver histology did not influence this comparison. The cumulative R2Y was 0.65, the cumulative Q2 was 0.37. In order to rule out that these results were due to overfitting, a seven-fold cross-validation and 300-fold permutation test was performed (see supplementary materials S1). The resulting regression lines showed a R2 intercept at 0.29 and a Q2 intercept at -0.39, indicating a valid model. This difference was maintained when the analysis of paired samples from primary and recurrent nodules was split. Identification of the metabolites which led to the difference between HCC and the adjacent parenchyma was performed using the S-plot analysis (Wiklund et al. 2008). Small metabolites statistically significant differentiating HCC and cirrhotic tissue were from signals at 0.86, 1.22, 1.26, 3.22, 3.26, 3.62, 3.9 ppm. According to BBIOREFCODE and HMDB database these signals correspond to CH3 fatty acid, (CH2)n fatty acids choline, trimethylamineoxide (TMAO), valine and glucose. This assignment was confirmed by the 2D TOCSY analysis. The univariate comparison of the normalized peak integrals of these signals (Table 2) shows that choline, TMAO and glucose were higher in HCC nodules than in the surrounding cirrhotic liver. When the analysis was extended including peaks derived from lipids, peaks at 0.86 (CH3 fatty acid), 1.22 ((CH2)n fatty acid), and 1.26 ((CH2)n fatty acid), significantly higher in the cirrhotic parenchyma, contributed to differentiate HCC from the adjacent tissue. 3.3 Metabolomics of primary and recurrent HCC The metabolite profile of primary and recurrent HCC lesions was compared by analyzing the spectra obtained from 12 samples obtained from primary HCC nodules and 16 from recurrent HCC lesions. By OPLS-DA there was a statistically significant difference between the two classes of tumors. The score plot of the predictive component vs the first of 5 orthogonal components (Fig. 3) did not show any overlapping. The cumulative R2Y was 0.84 and the cumulative Q2 was 0.21. There was no overfitting, as evaluated by a seven-fold cross-validation and

123

300-fold permutation in the corresponding PLS-DA with a R2Y intercept at 0.62 and a Q2 intercept at -0.63 (see supplementary materials S2). According to the S-plot analysis, the small metabolites statistically significant differentiating primary and recurrent HCC were from signals at 1.3, 1.34, 1.46, 2.02, 2.34, 3.18, 3.54, 3.62, 3.7 ppm. These signals correspond to (CH2)n fatty acid, lactate, alanine, proline, glutamate, phosphocholine, myo-inositol, valine and leucine. The univariate comparison of the normalized peak integrals of these signals (Table 2) showed that alanine, glutamate and myo-inositol were higher in primary HCC than in recurrences. Peak signals derived from lipids showed that CH3 fatty acid (at 1.3 ppm) also contributed to differentiation of primary from recurrent nodules. 3.4 Metabolomic analysis and tumor volume The relationship between the metabolite profile and the tumoral volume was evaluated by pooling samples from primary tumors and recurrencies and comparing the spectra obtained in tumors B2 cm (n = 15; diameter median 1.7 range 1.4–2) and in those larger than 2 cm (n = 11 diameter median 3.8 range 2.3–5). The score plot of the predictive component and the only orthogonal component (Fig. 4) shows distinct clusters of the 2 classes. This difference led to a model with R2Y 0.57 and Q2 0.22. There was no overfitting, as evaluated by a seven-fold cross-validation and 300-fold permutation in the corresponding PLS-DA with the R2Y intercept at 0.24 and the Q2 intercept at -0.16 (see supplementary materials S3). According to the S-plot analysis, the metabolites differentiating HCC of different volume were from signals at 1.34, 3.22, 3.38, 3.46, 3.82 corresponding to lactate, phosphocholine, b glucose, glucose and serine. However, the univariate comparison of the normalized peak integrals of the signals corresponding to these metabolites (Table 2) showed that these differences were marginal. By contrast, when the analysis was extended to lipids, we found that CH3 fatty acids and (CH2)n fatty acids were the metabolites of major interest differentiating HCC in relationship to the volume, as saturated fatty acids were significantly lower in large nodules. 3.5 HCC metabolomic analysis and diabetes The relationship between HCC profile and diabetes was also evaluated using diabetes as a distinct group. Although the metabolite profiles of parenchymal specimens from diabetic patients were different from those of non diabetic patients (data not shown), the comparison between nodules and adjacent tissue, as well the comparison of nodules from diabetic versus non diabetic patients did not yield significant differences.

HR-MAS NMR-based metabolomic fingerprinting Table 2 Metabolites differentiating HCC nodules and parenchyma, HCC primary from recurrent and nodules B2 cm from nodules [2 cm Metabolites differentiating HCC nodules and parenchyma ppm

Assignment

S-plot analysis Covariance

Univariate analysis p(corr)1

Nodule

Parenchyma

p

0.86

CH3 fatty acid

0.42

0.45

0.728

1.95

0.011

1.22

(CH2)n fatty acid

0.06

0.45

0.137

0.291

0.044

1.26

(CH2)n fatty acid

0.56

0.41

1 084

2.48

0.023

3.22

Choline

-0.61

-0.35

8.99

5.36

0.01

3.26

TMAO

-0.18

-0.35

2.92

1.43

0.007

3.62

Valine

0.17

0.42

2.53

3.29

0.017

3.9

Glycogen

-0.12

-0.33

3.06

2.40

0.022

Metabolites differentiating primary from recurrent HCC ppm

Assignment

S-plot analysis

Univariate analysis

Covariance

p(corr)1

Denovo

Recurrence

p

6.45

7.52

0.010

1.3

(CH2)n fatty acid

-0.23

-0.08

1.34

Lactate

-0.36

-0.29

1.46

Alanine

0.09

0.26

2.02

Proline

0.11

0.55

2.34

Glutamate

0.17

0.43

3.18

Phosphocholine

0.62

0.38

6.22

3.54

Myo-inositol

0.24

0.32

4.91

3.65

0.026

3.62

Valine

0.11

0.21

2.83

2.54

NS

3.7

Leucine

-0.19

-0.35

3.64

4.03

NS

0.991

2.59

NS

1.15

0.891

0.016

1.16

0.798

NS

1.97

1.15

0.016

5.67

NS

Metabolites differentiating nodules B2 cm from nodules [2 cm ppm

Assignment

S-plot analysis Covariance

Univariate analysis p(corr)1

B2 cm

[2 cm

p 0.015

0.86

CH3 fatty acid

0.20

0.74

1.10

0.362

0.9

CH3 fatty acid

0.50

0.68

6.40

1.99

0.025

1.26 1.3

(CH2)n fatty acid (CH2)n fatty acid

0.28 0.58

0.88 0.67

1.29 7.52

0.387 5.79

0.011 0.038

1.34

Lactate

0.22

0.53

1.58

0.913

NS

3.22

Choline

-0.39

-0.50

7.80

9.77

NS

3.38

b-Glucose

-0.14

-0.70

2.20

3.36

NS

3.46

Glucose

-0.17

-0.52

2.84

3.28

NS

3.82

Serine

-0.08

-0.70

2.89

3.36

0.04

NS not significative

3.6 Metabolite profiles in non tumoral tissue No difference was observed when the spectra of the liver parenchyma were compared according to the different classes of HCC nodules. 3.7 Discussion We found that HRMAS-NMR spectroscopy of minimal amounts (10–30 mg) of liver tissue and pattern recognition

analysis differentiate HCC from the surrounding parenchyma, primary HCC nodules from recurrences, and small from larger nodules. These results extend previous observations obtained in advanced tumors and indicate that small HCC nodules have distinct metabolite fingerprints. In our study the metabolomic findings were concordant with the radiological diagnoses, whereas histology, obtained by a single pass biopsy, failed to show the features of HCC in 5 of 15 primary HCC nodules and in 6 of 16 HCC recurrences. A possible explanation of this

123

A. Solinas et al. Fig. 3 Scores plot and S-plot resulting from OPLS-DA applied to the 1H HR-MAS NMR spectra of biopsies from recurrent (green circles) and primary (blue squares) HCC. In S-plot red circles refer to metabolites in Table 2 (Color figure online)

discrepancy is the tiny amount of tissue available for examination. Indeed, core biopsies of 5 mm obtained with a 18 gauge needle are sufficient for HR-MAS NMR spectroscopy, whereas a mean length of 15 mm is required to maximize the accuracy of the histological diagnosis. This finding implicates that NMR based metabolomics of

123

ultrasound guided needle biopsies could have a role in the diagnosis of very early HCC, when both histology and radiological imaging are less accurate than in large nodules (Iavarone et al. 2010). The metabolites differentiating HCC nodules from the surrounding cirrhotic liver were determined by S-plot

HR-MAS NMR-based metabolomic fingerprinting Fig. 4 Scores plot and S-plot resulting from OPLS-DA applied to the 1H HR-MAS NMR spectra of biopsies from nodules B2 cm (green inverted triangles) and [2 cm (red triangles). The maximum diameter of each nodule is expressed in cm. In S-plot red circles refer to metabolites in Table 2 (Color figure online)

analysis. The normalized peak areas of metabolites of relevance in the differentiating model were then compared by univariate analysis. The most relevant differences between tumoral and extranodal tissue were related to choline metabolism. High levels of choline, converted to phosphatidylcholine via the Kennedy pathway, is a hallmark of several neoplasms (Galons et al. 1995; Glund et al. 2011). High levels of TMAO, on the other hand, are the

product of phospholipid catabolism. These data are in agreement with the previous findings of Yang et al. (2007) and suggest that choline represents an early and reliable biomarker of HCC oncogenesis. By contrast, we did not observe an increased consumption of glucose, as the signals related to glucose were irrelevant in differentiating HCC from cirrhotic tissue, and signal at 3.9 ppm related to glucose and glycogen was higher in HCC specimens.

123

A. Solinas et al.

Whether this finding means that in the early phase of HCC glucose consumption is not increased remains to be clarified. Recent data emphasize that the rate of glycolysis in large HCC is significantly lower than in other neoplasms (Beyoglu et al. 2013). It is conceivable that in early HCC conversion to aerobic glycolysis is not the primary event in the metabolite changes related to oncogenesis. Finally, we found a relative increase in CH3 and (CH2)n fatty acids in the parenchyma in comparison to HCC nodules. Our data are in agreement with Yang et al. (2007). Beyoglu et al. (2013) also found that stearic acid and palmitic acid were virtually absent in the G1 transcriptomic group of HCC. In addition, dynamic MRI of HCC B2 cm detected the presence of intralesional fat only in 24 out of 159 (15.1 %) nodules (Rimola et al. 2012). By contrast, de novo lipogenesis has been described in human and in experimental HCC, (Calvisi et al. 2004; Budhu et al. 2013; Calvisi et al. 2011) and the clinical features of our patients do not fit the G1 trascriptomic group described by Boyault et al. (2007). Possibly, our finding should be interpreted taking into account the metabolic derangements of cirrhosis, characterized by a progressive increase of saturated fatty acids (Cobbold and Patel 2010). Moreover, the relationship with concomitant diabetes (which was diagnosed in 8 of 25 of our patients) could play a role. Our second aim was to evaluate whether metabolomics differentiates primary HCC from recurrences. Our results clearly indicate that these two classes of HCC have unique metabolite profiles. These findings, if validated in a larger cohort of patients, open to prognostic evaluation of early HCC based on needle biopsies. Higher lactic acid and lower myo-inositol indicate that recurrent HCC nodules are characterized by an increased glycolysis in comparison to primary HCC. Given the short interval between primary lesion and recurrence, it is conceivable that our series consisted of true metachronous metastases. The tissue of the primary nodules of these cases was not available for this study. The longitudinal analysis of primary tumors in patients free of progression compared to patients with recurrencies would clarify whether primary aggressive tumors characterized have also a distinct metabolomic profile. Finally, we addressed the issue of the metabolomic profile in relationship to the volume of the neoplasia. Although the model was weak, we have shown that HCC nodules tend to cluster according to their diameter. Interestingly, the diameter of the nodules was inversely related to saturated fatty acids. Yang et al. (2007) have reported the same finding in high grade HCC, and Budhu et al. (2013) have reported that the activity of stearolyl-CoAdesaturase is increased in aggressive HCC. These data suggest that a decrease in saturated fatty acid is a prognostic indicator of clinical relevance.

123

4 Conclusions Our study shows that metabolomics, based on HR-MAS NMR spectroscopy of needle biopsies, is a suitable tool in the diagnosis of early HCC and in the evaluation of its progression. Alterations in choline metabolism represent an early marker of HCC oncogenesis, whereas increased cytosolic glycolisis characterizes recurrent nodules. Neoplastic growth and saturated fatty acids are inversely related. Acknowledgments The authors wish to thank Dr. F. Benevelli, Dr. A. Minoja and Dr. C. Napoli (Bruker Italy) for suggestions and discussions and Dr. L. Barberini (University of Cagliari) for useful discussions.

References Altekruse, S. F., McGlynn, K. A., Dickie, L. A., & Kleiner, D. E. (2012). Hepatocellular carcinoma confirmation, treatment, and survival in surveillance, epidemiology, and end results registries, 1992–2008. Hepatology, 55(2), 476–482. Beyog˘lu, D., Imbeaud, S., Maurhofer, O., et al. (2013). Tissue metabolomics of hepatocellular carcinoma: Tumor energy metabolism and the role of transcriptomic classification. Hepatology,. doi:10.1002/hep.26350. Boyault, S., Rickman, D. S., de Reynie`s, A., et al. (2007). Transcriptome classification of HCC is related to gene alterations and to new therapeutic targets. Hepatology, 45(1), 42–52. Budhu, A., Roessler, S., Zhao, X., et al. (2013). Integrated metabolite and gene expression profiles identify lipid biomarkers associated with progression of hepatocellular carcinoma and patient outcomes. Gastroenterology, 144(5), 1066–1075. Bylesjo¨, M., Rantalainen, M., Cloarec, O., Nicholson, J. K., Holmes, E., & Trygg, J. (2006). OPLS discriminant analysis: combining the strengths of PLS-DA and SIMCA classification. Journal of Chemometrics, 20, 341–351. Calvisi, D. F., Factor, V. M., & Ladu, S. (2004). Disruption of betacatenin pathway or genomic instability define two distinct categories of liver cancer in transgenic mice. Gastroenterology, 126, 1374–1386. Calvisi, D. F., Wang, C., Ho, C., et al. (2011). Increased lipogenesis, induced by AKT-mTORC1-RPS6 signaling, promotes development of human hepatocellular carcinoma. Gastroenterology, 140(3), 1071–1083. Chun, J. M., Kwon, H. J., Sohn, J., et al. (2011). Prognostic factors after early recurrence in patients who underwent curative resection for hepatocellular carcinoma. Journal of Surgical Oncology, 103(2), 148–151. Cobbold, J. F. L., & Patel, J. H. (2010). Hepatic lipid profling in chronic hepatitis C: an in vitro and in vivo proton magnetic resonance spectroscopy study. Journal of Hepatology, 52, 16–24. Forner, A., Llovet, J. M., & Bruix, J. (2012). Hepatocellular carcinoma. Lancet, 379, 1245–1255. Galons, J. B., Job, C., & Gilles, R. (1995). Increase of GPC levels in cultured mammalian cells during acidosis. Magnetic Resonance in Medicine, 33, 422–426. Glund, K., Bhujiwalla, Z. M., & Rohen, S. M. (2011). Choline metabolism in malignant transformation. Nature Reviews Cancer, 11, 835–848.

HR-MAS NMR-based metabolomic fingerprinting Gogvadze, V., Zhivotovski, B., & Orrenius, S. (2009). The Warburg effect and mitochondrial stability in cancer cell. Molecular Aspects of Medicine, 31, 60–74. Griffin, J. L., & Shockcor, J. P. (2004). Metabolic profiles of cancer cells. Nature Reviews Cancer, 4, 551–561. Iavarone, F., Sangiovanni, A., & Forzenigo, L. V. (2010). Diagnosis of hepatocellular carcinoma in cirrhosis by dynamic contrast imaging: the importance of tumor cell differentiation. Hepatology, 52(5), 1723–1730. Livraghi, T., Meloni, F., Di Stasi, M., et al. (2008). Sustained complete response and complications rates after radiofrequency ablation of very early hepatocellular carcinoma in cirrhosis: Is resection still the treatment of choice? Hepatology, 47(1), 82–89. Lo´pez-La´zaro, M. (2008). The Warburg effect: Why and how do cancer cells activate glycolysis in the presence of oxygen? Anticancer Agents in Medicinal Chemistry, 8, 305–312. Martinez-Granados, B., Monleon, D., Martinez-Bisbal, M. C., et al. (2006). Metabolite identification in human liver needle biopsies by high-resolution magic angle spinning 1H NMR spectroscopy. NMR in Biomedicine, 19, 90–100. Minquez, B., Hoshida, Y., & Villanueva, A. (2011). Gene-expressione signature of vascular invasion in hepatocellular carcinoma. Journal of Hepatology, 55(6), 1325–1331. Negrini, S., Gorgoulis, V. G., & Halozenetis, T. D. (2010). Genomic instability- an evolving hallmark of cancer. Nature Reviews Molecular Cell Biology, 11, 220–228. Ng, D. J. Y., Pasikanti, K. K., & Chan, E. C. Y. (2011). Trend analysis of Metabolomics and systematic review of metabolomicsderived cancer marker metabolites. Metabolomics, 7, 155–178. Paris, D., Melck, D., Stocchero, M., D’Apolito, O., Calemma, R., Castello, G., et al. (2010). Monitoring liver alterations during

hepatic tumorigenesis by NMR profiling and pattern recognition. Metabolomics, 6, 405–416. Park, Y., Choi, D., Lim, H. K., et al. (2008). Growth rate of new hepatocellular carcinoma after percutaneous radiofrequency ablation: evaluation with multiphase CT. American Journal of Roentgenology, 191(1), 215–220. Rimola, J., Forner, A., Tremosini, S., et al. (2012). Non invasive diagnosis of hepatocellular carcinoma B2 cm in cirrhosis. Diagnostic accuracy. Journal of Hepatology, 56, 1317–1323. Roayaie, S., Obeidat, K., Sposito, C., et al. (2013). Resection of hepatocellular cancer B2 cm: Results from two western centers. Hepatology,. doi:10.1002/hep.25832. Sangiovanni, A., Del Ninno, E., Fasani, P., et al. (2004). Increased survival of cirrhotic patients with a hepatocellular carcinoma detected during surveillance. Gastroenterology, 126, 1005–1014. Trygg, J., & Wold, S. (2002). Orthogonal projections to latent structures, O-PLS. Journal of Chemometrics, 16, 119–128. Wiklund, S., Johansson, E., Sjo¨stro¨m, L., et al. (2008). Visualization of GC/TOF-MS-based metabolomics data for identification of biochemically interesting compounds using OPLS class models. Analytical Chemistry, 80(1), 115–122. Wishart, D. S., Jewison, T., Guo, A. C., et al. (2013). HMDB 3.0— The human metabolome database in 2013. Nucleic Acids Research, 41(Database issue), D801–D807. Yang, Y., Li, C., Nie, X., & Feng, X. (2007). Metabonomic studies of human hepatocellular carcinoma using high resolution magic angle spinning. Journal of Proteome Research, 6, 2605–2614. Zhang, B., Yang, B. H., & Tang, Z. Y. (2004). Randomized controlled trial of screening for hepatocellular carcinoma. Journal of Cancer Research and Clinical Oncology, 130(7), 417–422.

123