Rapid Diagnosis and Staging of Colorectal Cancer Via High-resolution Magic Angle Spinning Nuclear Magnetic Resonance (HR-MAS NMR) Spectroscopy of Intact Tissue Biopsies

Reza Mirnezami, MBBS, MRCS; Beatriz Jiménez, PhD, MRSEQ, MSEBBM; Jia V. Li, PhD; James M. Kinross, PhD, FRCS; Kirill Veselkov, PhD; Robert D. Goldin, MD, FRCPath; Elaine Holmes, PhD, FRSC; Jeremy K. Nicholson, PhD, FRCPath, FMedSci; Ara Darzi, MD, FRCS, FRCP, FMedSci, FRS

Disclosures

Annals of Surgery. 2014;259(6):1138-1149.

Abstract and Introduction

Abstract

Objective. To develop novel metabolite-based models for diagnosis and staging in colorectal cancer (CRC) using high-resolution magic angle spinning nuclear magnetic resonance (HR-MAS NMR) spectroscopy.

Background. Previous studies have demonstrated that cancer cells harbor unique metabolic characteristics relative to healthy counterparts. This study sought to characterize metabolic properties in CRC using HR-MAS NMR spectroscopy.

Methods. Between November 2010 and January 2012, 44 consecutive patients with confirmed CRC were recruited to a prospective observational study. Fresh tissue samples were obtained from center of tumor and 5 cm from tumor margin from surgical resection specimens. Samples were run in duplicate where tissue volume permitted to compensate for anticipated sample heterogeneity. Samples were subjected to HR-MAS NMR spectroscopic profiling and acquired spectral data were imported into SIMCA and MATLAB statistical software packages for unsupervised and supervised multivariate analysis.

Results. A total of 171 spectra were acquired (center of tumor, n = 88; 5 cm from tumor margin, n = 83). Cancer tissue contained significantly increased levels of lactate (P < 0.005), taurine (P < 0.005), and isoglutamine (P < 0.005) and decreased levels of lipids/triglycerides (P < 0.005) relative to healthy mucosa (R²Y = 0.94; Q²Y = 0.72; area under the curve, 0.98). Colon cancer samples (n = 49) contained higher levels of acetate (P < 0.005) and arginine (P < 0.005) and lower levels of lactate (P < 0.005) relative to rectal cancer samples (n = 39). In addition unique metabolic profiles were observed for tumors of differing T-stage.

Conclusions. HR-MAS NMR profiling demonstrates cancer-specific metabolic signatures in CRC and reveals metabolic differences between colonic and rectal cancers. In addition, this approach reveals that tumor metabolism undergoes modification during local tumor advancement, offering potential in future staging and therapeutic approaches.

Introduction

The mortality associated with colorectal cancer (CRC) remains unacceptably high despite the widespread adoption of multimodal treatment approaches that include surgery, chemoradiotherapy,^[1,2] and novel targeted agents in select cases.^[3] Relative 5-year survival remains below 60% in most European countries,^[4] and improving on this figure will require a more complete understanding of the fundamental molecular processes that drive CRC development and progression. During the past 2 decades, genomic and transcriptomic studies have identified several key molecular pathways in colorectal carcinogenesis, from which at least 5 distinct molecular phenotypes have emerged.^[5] The metabolome lies furthest downstream along the "-omics" cascade and offers an additional perspective on CRC biology. It is defined as the complete set of low-molecular-weight metabolites contained within a living system^[6,7] and represents a largely untapped biological resource in cancer research.

Recent advances in analytical biochemistry have led to the development of an array of metabolic profiling platforms with which to exploit the metabolome.^[7] In this study, we have applied high-resolution magic angle spinning nuclear magnetic resonance (HR-MAS NMR) spectroscopy^[8] combined with in-house developed biostatistical approaches to investigate the metabolic phenotype of CRC; this technique permits ultrafast, nondestructive tissue metabolic profiling, providing a unique sample-specific metabolic "fingerprint." A key advantage of this technology over other platforms is that sample preparation steps result in minimal tissue architectural disruption, meaning that sample tissue can be retrieved, post–HR-MAS NMR profiling, and subjected to conventional histopathological evaluation.^[8–10] Thus, HR-MAS NMR spectroscopy–derived metabolic phenotypes can be directly and precisely correlated with corresponding histological/immunohistochemical parameters of interest. This is in contrast to other metabolic profiling approaches such as gas chromatography mass spectrometry,^[11] where tissue samples are typically combined with complex solvent mixtures, ultrasonicated, centrifuged, and repeatedly vortex-mixed, rendering them unfeasible for subsequent histopathological assessment.

To date, a number of small studies have used HR-MAS NMR–based profiling for characterization of CRC metabolism.^[11–13] Chan et al^[11] described alterations in protein turnover, lipid composition, and glycolysis in CRC tissue compared with matched control samples. The authors sought to develop metabolite-based models for tumor stage classification but were unable to achieve this objective owing to the small number of samples in each stage class. Righi et al^[13] published HR-MAS NMR–based data on 23 patients (14 patients with CRC and 9 healthy subjects) and reported similar metabolic characteristics associated with CRC. This study was somewhat limited by the small sample size, and in addition the authors did not histologically evaluate HR-MAS NMR profiled tissue samples, leading to uncertainty regarding exact tissue composition.^[13] Our group has more recently published data on a previously recruited cohort of patients (no overlap with this study population) with CRC tissue samples and healthy adjacent mucosa (Hm) evaluated by HR-MAS NMR spectroscopy; this study demonstrated the potential for prediction of disease-free survival after CRC resection using metabolite-based models.^[12]

Building on previous work, this study was undertaken to address 3 specific objectives: (1) to comprehensively characterize the CRC metabolome; (2) to determine the metabolic differences between colonic and rectal cancer tissue; and (3) to determine whether HR-MAS NMR spectroscopic profiles can be used to predict local tumor extent (T-stage). Study objectives, methodology, results, and key conclusions are set out in accordance with guidelines proposed by the Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK) framework,^[14–16] and these components are summarized in the REMARK checklist provided ( Table 1 ).

1 2 3 4 5

Next Section

Abstract and Introduction
Methods
Results
Discussion
Conclusions
References

Table 1. REMARK Checklist.
Introduction
Marker(s) examined	HR-MAS NMR spectroscopy–derived metabolic biopanels from intact tissue samples have been used to identify differences at metabolic level between (i) cancerous and healthy colorectal mucosa, (ii) colon and rectal cancer mucosa, and (iii) tumors of different T-stage.^*
Hypotheses	(i) HR-MAS NMR metabolic profiles can reliably distinguish between cancerous and healthy colorectal mucosa.
	(ii) There are local tissue metabolic differences between colon and rectal tumor tissue.
	(iii) Tumors of differing T-stage will exhibit different local metabolic phenotypes.
Study objectives	To develop sensitive and specific HR-MAS NMR metabolic profiling strategies for discriminating: (i) cancerous from healthy colorectal mucosa, (ii) colon and rectal tumor tissue, and (iii) tumors of differing T-stage.
Materials and methods
Patient recruitment	After exclusion of 20 patients (reasons outlined below) 44 consecutively recruited adult patients (>18 yr) undergoing elective resection for biopsy confirmed CRC at a single center (St Mary's Hospital, London, United Kingdom) were included in the study. Reasons for study exclusion: (i) Enrolled to FOxTROT study (n = 7), (ii) Declined participation (n = 4), (iii) Lack of pathology personnel for evaluation and sampling of specimen (n = 4), and (iv) Inability to sample rectal cancer specimen due to potential compromising of CRM assessment status (n = 5)
Specimen characteristics	Fresh samples of colorectal mucosa retrieved from (i) center of tumor and (ii) 5cm from tumor margin from all surgical resection specimens.
Assay methods	HR-MAS NMR experiments performed using a Bruker Avance III 400-MHz spectrometer equipped with robotic autosampler. Spin rate set to 4500 Hz.
Study design	Prospective observational study
Statistical methods used	Unsupervised (PCA) and supervised (PLS-DA) multivariate analysis of data using SIMCA and MATLAB statistical software. For PCA 3 × 3 bi-cross-validation was used. For PLS-DA we performed 7-fold cross-validation and diagnostic model accuracy was assessed using Q ² statistic (measure of model predictive capacity) and ROC curve analysis on cross-validated (predicted) discriminatory scores.
Discriminatory metabolic biopanel determination	The discriminatory scores for separating samples according to predefined hypotheses were derived as a weighted sum of all measured metabolic features. However, the majority of metabolic features have negligible contribution to the model (zero weight) and therefore we have reported only those exhibiting the highest influence (ie, weight) on the model.
Results
Model-specific metabolic biopanel composition	(i) Cancerous vs healthy colorectal mucosa (cancer relative to healthy): ↑ lactate ↑ isoglutamine ↑ GPC ↑ taurine ↑ scyllo-inositol ↑ glycine; ↓ lipids/triglycerides ↓ α-glucose ↓ β-glucose
	(ii) Colon vs rectal cancer (colon relative to rectum): ↑ acetate↑ arginine; ↓ lactate
	(iii) T1–T 2 vs T3 (T1–T2 relative to T3): ↑ GPC; ↓ lipids/triglycerides ↓ acetate T3 vs T4 (T3 relative to T4) ↑ succinate ↑ acetate ↑ lipids/triglycerides
Biostatistical analysis	(i) Cancerous vs healthy colorectal mucosa: R ² Y = 0.94; Q ² Y = 0.72; 7-fold cross-validated ROC curve analysis on discriminatory scores AUC = 0.98
	(ii) Colon vs rectal cancer: R ² Y = 0.94; Q ² Y = 0.33; 7-fold cross-validated ROC curve analysis on discriminatory scores AUC = 0.86
	(iii) T-stage: 7-fold cross-validated classification accuracy of metabolite-based model to correctly identify tumors as T1–T2 = 91%, T3 = 90%, and T4 = 75%
Discussion
Interpretation of results	(i) HR-MAS NMR profiling reveals distinct cancer-associated metabolic features with dysregulated glycolytic mechanisms, lipidomic activity and protein turnover
	(ii) Colon and rectal cancer tissue harbor different metabolic signatures
	(iii) CRC microenvironmental metabolism undergoes modification during local tumor advancement
Clinical application	(i) Building enhanced appreciation of CRC metabolism for identification of novel anticancer drug targets
	(ii) Improving current understanding of molecular heterogeneity between tumors of the colon and rectum, which can help to explain observed phenotypic differences
	(iii) Stage-specific metabolic profiles to supplement conventional local staging strategies and design novel therapies based on stage-specific disruption of tumor metabolism
Future research	(i) Large-scale validation of findings presented in this study
	(ii) Identification of tissue metabolic characteristics associated with response to NA therapy
	(iii) Prospective comparison of predictive capacity of HR-MAS NMR profiling for local stage determination compared with established gold standards (CT and MRI)

*Pathological determination of T-stage according to the fifth edition of the UICC Tumor-Node-Metastasis (TNM) classification.
CRM indicates circumferential resection margin; PCA, principal component analysis; PLS-DA, partial least squares–discriminant analysis; GPC, glycerophosphoryl-choline.

Table 2. Demographic and Clinicopathological Characteristics
	N	%
Age, median (range)	72 (56–87)
Sex
Male	26	59
Female	18	41
Location of primary tumor
Transverse colon	3	7
Descending/proximal sigmoid colon	5	11
Ascending colon	6	14
Rectosigmoid	8	18
Cecum	10	23
Rectum	12	27
Tumor differentiation
Well-differentiated	1	2
Moderately differentiated	32	73
Poorly differentiated	11	25
pT stage
T1	3	7
T2	9	20
T3	20	46
T4	12	27
UICC stage
I	9	20
II	18	41
III	15	34
IV	2	5
Lymph node status
N0	27	61
N1/N2	17	39
EMVI
Yes	18	41
No	26	59
LVI
Yes	14	32
No	30	68
PNI
Yes	7	16
No	37	84
Preoperative chemotherapy/radiotherapy
LCCR	3	7
SCR	4	9
None	37	84

LCCR indicates long-course chemoradiotherapy; SCR, short-course radiotherapy; EMVI, extramural vascular invasion; LVI, lymphovascular invasion; PNI, perineural invasion.

Table 3. Summary of NMR Spectral Metabolite Assignments With Corresponding Chemical Structures, Chemical Shifts (ppm) and Observed Pattern Variations Between Cancerous and Healthy Colorectal Mucosa
Peak No.*	Metabolite	Chemical group	ppm	Relative pattern intensity	Function
1	Lipids/triglycerides	-CH₃	0.90	↑ in healthy	Cell membrane biosynthesis
2	Lipids/triglycerides	-(CH₂)_n-	1.29	↑ in healthy	Cell membrane biosynthesis
3	Lipids/triglycerides	C=C-CH₂	2.02	↑ in healthy	Cell membrane biosynthesis
4	Lipids/triglycerides	=CH-	5.33	↑ in healthy	Cell membrane biosynthesis
5	Lactate	-CH₃	1.33	↑ in cancer	By-product of ↑ cancer cell glycolysis
6	Lactate	-CH-	4.15	↑ in cancer	By-product of ↑ cancer cell glycolysis
7	Valine	-CH₃	1.02	↔	Ketogenic amino acid
8	Alanine	-CH₃	1.47	↔	Gluconeogenic amino acid
9	Leucine	β-CH₂ γ -CH	1.72	↔	Ketogenic amino acid
10	Acetate	-CH₃	1.92	↔	By-product of gut microbial fiber fermentation
11	Isoglutamine	β-CH₂	2.34	↑ in cancer	By-product of gut microbial cell wall degradation
12	Creatine	-CH₃	3.02	↔	Enhancement of ATP regeneration
13	Creatine	-CH₂-	3.93	↔	Enhancement of ATP regeneration
14	Glycerophosphorylcholine (GPC)	N(CH₃)₃	3.22	↑ in cancer	Cell volume regulation
15	Taurine	-CH₂-NH	3.26	↑ in cancer	Protein-building block; cell membrane stabilization
16	Taurine	CH₂SO₃	3.42	↑ in cancer	Protein-building block; cell membrane stabilization
17	Scyllo-inositol	-O-H	3.34	↑ in cancer	Osmotic regulation
18	Glycine	CH₂	3.56	↑ in cancer	Protein-building block
19	β-glucose	1-CH	4.67	↑ in healthy	Energy generation
20	α-glucose	1-CH	5.24	↑ in healthy	Energy generation