Applications of High-Resolution Magic Angle Spinning MRS in Biomedical Studies II -- Human Diseases

Abstract

High-resolution magic angle spinning (HRMAS) magnetic resonance spectroscopy (MRS) is a powerful method for gaining insight into the physiological and pathological processes of cellular metabolism. Given its ability to obtain high resolution spectra of non-liquid biological samples, while preserving tissue architecture for subsequent histopathological analysis, the technique has become invaluable for biochemical and biomedical studies. Using HRMAS MRS, alterations in measured metabolites, metabolic ratios, and metabolomic profiles present the possibility to improve identification and prognostication of various diseases and decipher the metabolomic impact of drug therapies. In this review, we evaluate HRMAS MRS results on human tissue specimens from malignancies and non-localized diseases reported in the literature since the inception of the technique in 1996. We present the diverse applications of the technique in understanding pathological processes of different anatomical origins, correlations with in vivo imaging, effectiveness of therapies, and progress in the HRMAS methodology.

Graphical Abstract

High resolution magic angle spinning (HRMAS) magnetic resonance spectroscopy (MRS) is a method which enhances resolution for non-liquid biological samples. Since its introduction in 1996, HRMAS has been used in a wide variety of biomedical and biochemical studies and is one of the major techniques used in metabolomic analysis. In this review, HRMAS results on human tissue malignancies and non-localized diseases are evaluated, with discussions of applications for understanding pathological processes, correlations with in vivo imaging, therapy effectiveness, and methodology progress.

Introduction

The utility of magnetic resonance spectroscopy (MRS) as a medical diagnostic, prognostic and therapeutic evaluation tool for a wide range of diseases has been well investigated. MRS, particularly high-resolution magic angle spinning (HRMAS) MRS, has demonstrated its ability for direct measurement of non-liquid biological tissue to obtain valuable insight into cellular metabolism of physiological and pathological processes of human diseases, as well as into experimental models of animals and cell-lines of these medical conditions.

In this article, we aim to comprehensively review HRMAS MRS for its applications in ex vivo studies of diseased human tissue specimens published prior to the end of 2016. Hereinafter, we narrowly define human studies as direct HRMAS MRS measurements of human tissue specimens, not including animal and cell-line studies of human disease models, as they were reviewed in a separate article¹.

HRMAS MRS observable tissue spectra present mostly cellular metabolites and mobile lipids. Metabolites are the signatures of biochemical pathways, through which the processes of cellular carbohydrate, protein, lipid, amino acid and xenobiotic metabolism may be probed. The ability of MRS to explore the metabolic status can help elucidate the physiological and pathological transformations accompanying various medical conditions; it can provide a detailed account of metabolic mechanisms at cellular levels^2–4. Review of the published HRMAS MRS studies on human disease emphasizes the utility of the technique especially in understanding human malignancies. Cancer presents identifiable lesions, which enables clear comparison of metabolites from affected and not affected tissue from the same patient. For non-malignant medical conditions, metabolite investigation is more complicated. Disease and control tissues must be obtained from different individuals, presenting bio-variation as a confounding factor. Therefore, the large body of HRMAS MRS research is skewed towards cancer, but we will balance our presentation with critical reviews of other non-localized diseases.

Applications of HRMAS MRS on studies of human malignant diseases

Publications of HRMAS MRS human studies on cancers revealed areas of research emphases that reflect either the prevalence of diseases seen in clinic or the technical advancement achieved with clinical MRS. This review analyzed over 100 studies using HRMAS MRS, including brain tumors (41 papers), breast cancer (21), prostate cancer (19), cervical cancer (9), gastrointestinal tumors (9), lung cancer (5) and renal cancer (4). Cancers are highly variable as far as patient prognostication is concerned. The major aims of these studies, in addition to providing metabolic markers for disease detection and diagnosis, were centered on identification of disease aggressiveness and their response to systemic therapies.

Breast and prostate cancers, as the most frequently diagnosed cancers respectively for women and men, respectively, with 252,710 and 161,360 estimated new cases and accounting for 40,610 and 26,730 estimated deaths in the U.S. in 2017 alone⁵, have attracted the most attention in research. On the other hand, although tumors of the brain and central nervous system might only present 23,800 new cases and account for 16,700 deaths⁵, two specific characteristics of the field encouraged researchers to focus on these malignancies: the unparalleled advancement of brain MR techniques (imaging, MRI; functional imaging, fMRI, and MRS) and the vast number of clinical studies reported in the area, due mostly to the relative tissue magnetic susceptibility homogeneity, and minimal respiratory and cardiac motion effects in the brain.

Brain tumors

To structure a large body of publications in this area^6–52 into a differentiated review, we will divide this section into three focuses: the correlations between in vivo and ex vivo HRMAS MRS, classification of adult and pediatric CNS tumors, and the relationships between metabolites and histopathological tissue properties. As “brain tumors” refer to a large class of neoplasia including both malignant and non-malignant lesions, we aimed to put the different metabolic findings into the context of the pathologic grading (I–IV) according to the 2016 WHO brain tumor classification⁵³, to reflect the tumor differentiation levels. As the most frequent human brain tumor, most of the research was targeted at gliomas and their subtypes, e.g. astrocytomas and oligodendrogliomas, with the pediatric medulloblastoma also the focus of several studies.

In vivo and ex vivo correlations

While clinical MR investigations on brain tumors have inspired studies of ex vivo HRMAS MRS of tissue specimens to reveal tumor type-specific metabolic structures and pathways, the most significant value of such ex vivo tissue analyses is in improving interpretation of different tumor types, grades, and progression potential for in vivo MRS. A study of 17 human astrocytomas (4 grade II, 1 grade III and 12 grade IV) has shown spectral similarities between in vivo and ex vivo results (Figure 1), and highly significant correlations were found for 8 out of the 13 measured metabolite concentrations, including Creatine (Cr), Glutmate (Glu), Glutamine (Gln), myo-inositol (myo-Ino), N-acetylaspartate (NAA), scyllo-inositol (scy-Ino), total choline (tCho), Glycerophosphocholine (GPC) and Phosphocholine (PCho)) and Lipid/Macromolecule (Lip/MM)²⁶. Similarities between in vivo and ex vivo MRS have also been reported for heterogeneous primary and secondary glioblastomas (GBM)¹⁹, as well as for hemangiopericytoma and meningiomas, two less heterogeneous tumor types³¹.

Figure 1.

In vivo PRESS TE 30 msec spectra (b, e, h) and their equivalent ex vivo HRMAS presaturation spectra (c, f, i) from a histopathologically verified astrocytoma grade II (a–c) and glioblastoma (d–f) showing unimodal variation of the grayscale pixel values in the voxel placement areas (a,d), and a histologically verified glioblastoma with multimodal variation of the grayscale pixel values (g–i). Major peaks of the metabolites discussed have been labeled, but for clarity not all metabolites are labeled in each spectrum (Ala, alanine; Cr, creatine; Gln, glutamine; Glu, glutamate; Glx, [Gln + Glu]; Gly, glycine; Ins, myo-inositol; Lac, lactate; Lip/MM, lipids/macromolecules; Scyllo, scyllo-inositol; Tau, taurine; tCho, total cholines)²⁶.

However, when connecting ex vivo HRMAS with in vivo MRS results, the major difference between the two spectra, due partially to the magnetic field strength difference and the application of HRMAS in the ex vivo measurements, needs to be considered, particularly when concerning the overlap among metabolite signals in the in vivo MRS. For instance, in the last mentioned hemangiopericytoma and meningiomas study, mannitol – a medication used to reduce brain pressure – presented one peak in the 3T in vivo spectrum at 3.8 ppm. However, when measured ex vivo at 9.4T, the HRMAS spectrum showed multiple peaks between 3.6 and 4.0 ppm. This phenomenon is common for overlapping signals of various metabolites, such as for GPC, PCho, and choline (Cho), and Glu and Gln, which leads the former to be assigned as total choline (tCho = GPC+PCho+Cho), and the latter as Glx (Glu+Gln) for in vivo MRS reports²⁶.

Differentiations among individual metabolites can be very important for tumor identification as noted in a comparative study between grade II astrocytomas (n=10) and GBMs (n=48), where PCho is increased in grade II astrocytomas, while GBMs showed increased levels of GPC. In addition, different amounts of GPC were seen to significantly separate primary GBMs from secondary GBMs (p<0.001). Therefore, accurate differentiations between PCho and GPC, rather than considering tCho collectively, may make it possible to distinguish between grade II astrocytomas and GBM and within the GBM subtype⁴⁰.

In addition to metabolite overlaps, the detection of certain metabolites can also be field strength- and sensitivity-dependent. For instance, alanine (Ala) was detectable in an HRMAS spectrum at 9.4T but not at 3T during in vivo MRS³¹. Despite these technical challenges, the trend of increasing magnetic field strength of clinical MRI scanners means the clinical translation of HRMAS MRS findings is increasingly possible. Currently, HRMAS MRS is highly suitable to tackle another task that presents challenges in the clinical setting: analysis of tissue metabolic structures to characterize and distinguish brain tumor types and grades.

Differentiations of tumor types and grades for adult and pediatric brain cancer

Due to significant clinical impact, gliomas have been a primary focus of research in the recent years. In 2010, an HRMAS MRS study of 30 neuroepithelial brain cancer specimens showed that certain metabolic changes are directly or inversely related to tumor grades (Figure 2). Specifically, in high grade tumors (grade III–IV), taurine (Tau), glycine (Gly) and aspartate (Asp) increases and Cr, myo-Ino, GPC, PCho decreases were observed. Furthermore, lipids and Ala were higher in grade II and III astrocytomas than in grade I. For grade III astrocytoma, NAA, Cr, myo-Ino (all p<0.01) were lower than in grades I and II astrocytomas, whereas Gly (p<0.05) was higher. GBM spectra were mostly characterized by lower Lac, Cr, and myo-Ino (all p<0.05) than in grades I–III astrocytomas. To distinguish grade III astrocytomas from GBM, GPC (n.s.), Cho (n.s.), lactate (Lac) (p<0.05) and Cr (p<0.001) are found in higher levels in the former. In contrast, myo-Ino (p<0.001) and PCho (n.s.) are lower in grade III. The changes in the following five metabolites were in agreement with other in vivo studies: Gly, NAA, Lac, Cr, and myo-Ino⁸. A summarizing table of reported brain tumor HRMAS MRS measurable metabolite changes from this and other reviewed references will be provided later in this section.

Figure 2.

HR-MAS ¹H NMR spectra of six brain specimens acquired from CPMG pulse sequence at a spin rate of 2 kHz. (A) Normal Tissue, (B) Astrocytoma (grade I), (C) Astrocytoma (grade II), (D) Astrocytoma, (E) Glioblastoma (grade IV), (F) Medulloblastoma (grade IV)⁸.

In another study, metabolic differences between grade II astrocytomas (n=10) and GBMs (n=48) were investigated using multivariate and principal component analysis (PCA). GBMs showed increased lipids and decreased GPC levels than grade II astrocytomas, which accords with previous research, in addition to harboring higher PCho and Gly and lower myo-Ino. The prediction model showed a sensitivity of 70% and a specificity of 88% in discriminating grade II from GBMs. As previously described, primary and secondary GBMs can be differentiated by PCho, with recurrent GBM (n=17 out of the 48) having lower amounts of this metabolite⁴⁰.

Gliomas tend to not only recur but also transform to a higher grade, so biomarkers that can identify these types can be crucial in the clinic. Differentiation between grade II gliomas that underwent malignant transformation to a higher grade vs. those that remained grade II was evaluated for 53 recurrent grade II gliomas. Myo-Ino levels were 56% lower in gliomas that underwent progression than those that remained low grade, while 2-hydroxyglutarate (2HG) (2.24 – 2.30 ppm), hyp-Tau (2.62–2.69 ppm), and Cho (3.21–3.29 ppm) had up to 120%, 137%, and 83% higher levels in the progressive gliomas. Higher levels in this group were also detected for lipids, gluthathione (GSH) and Ala. Particularly, myo-Ino was proposed as a key metabolite of discriminative strength. Combined with other presented metabolic information, it assisted in reaching 96% accuracy in identifying tumor progression¹⁰.

Later in 2014, a larger study with 254 samples from 126 patients with either primary or secondary and new or recurrent gliomas was reported. Comparison between primary GBMs (n= 101) and secondary GBMs (n_total=33, grade II → grade IV: 19, grade III → IV: 14) found that tCho, 2HG and Asp were significantly higher in the grade II → IV cases than in primary GBMs (all p<0.005). In addition, classification models could separate primary from secondary GBMs with very high accuracy, scoring 94% between grade II → IV and 96% between grade III → IV, with 2HG being the most important parameter. Separating these types is vital since GBMs with transformative potential need more aggressive treatment than de novo GBMs^46–48,51. Primary grade II and grade III astrocytomas could be separated from primary GBMs with higher 2HG levels for the astrocytomas. Additionally, a higher myo-Ino/tCho ratio and a lower total glutathione (tGSH = reduced glutathione + disulfide form) level were present in grade II tumors vs. GBMs, and higher levels of tCho, 2HG, NAA, Cr/phosphocreatine (PCr) ratio existed for the grade III to enable separation from the GBMs. Astrocytomas themselves could be separated with a higher myo-Ino/tCho ratio, together with lower levels of Gly and phosphoethanolamine (PE), distinguishing primary grade II from primary grade III anaplastic astrocytomas with 73% accuracy¹⁵.

The tumor-promoting onco-metabolite 2HG⁵², as an important discrimination marker for gliomas, was found to be significantly correlated with a number of other metabolites in a 2009 study. Positive correlations were reported for fifteen metabolites [PCho, GPC, tCho, Cr, phosphocreatine (PCr), PE, GSH, Glu, Gln, gamma-aminobutyric acid (GABA), Ala, Gly, betaine (Bet), hypo-taurine (hyp-Tau), and Lac], and 2HG was negatively correlated with the myo-Ino/tCho ratio (all p<0.05). Choline-containing compounds (ChoCC) are widely used as an in vivo MRS marker for tumor presence and increased levels of many other metabolites, so 2HG could be an added marker. It could also contribute additional prognostic information in the clinical setting. For example, glioma patients with an IDH1 or IDH2 gene mutation present more favorable prognoses (higher 5-year survival rate) and accumulate higher amounts of 2HG⁵². Considering the resonance overlaps between 2HG and other metabolites at clinical field strengths¹⁴, ex vivo MRS identification of this relationship is critical. Positively, other findings suggest that 2HG can be identified clinically by using 1D spectral-editing with a localized 2D correlation⁷.

HRMAS MRS results have also been supported by genomic analyses. In a study of 9 grade II, 6 grade III, 8 grade IV astrocytomas and 3 normal tissue specimens, as expected, results found that NAA and GABA, two important neuronal markers, were decreased in glioma tissues as compared to normal tissue. Using higher GPC/tCho ratio in low grade (II) tumors, the latter can be correctly separated from high grade (III and IV) tumors in 22 out of 23 cases (95% accuracy), suggesting GPC as the main component in low grade tumors and PCho as a signature Cho derivate in high grade glioma. Genomic results further supported the purported rise in PCho by demonstrating an increase in the gene expressions of choline kinase α and phospholipase C²⁹.

Whereas most HRMAS MRS research on malignant disease focuses on adults, research on brain malignancies extends to pediatric studies. One of the most common malignant CNS tumors in children is the medulloblastoma^49,50, which is a high grade (WHO grade IV) tumor with an overall poor prognosis. A cohort of 35 cerebellar childhood tumors (18 medulloblastomas, 12 pilocytic astrocytomas and 5 ependymomas), were studied, with separation possible between all three tumor types. High levels of PCho and Glu as well as low Gln levels characterized medulloblastomas. These tumors also had high Tau, MM/Lipids and Ala levels. Alterations in NAA, Cr and myo-Ino can be used to distinguish ependymomas from pilocytic astrocytomas, with NAA detectable in astrocytomas, but not in ependymomas or medulloblastomas. Higher Cr levels are found in ependymomas (Figure 3)¹².

Figure 3.

Mean spectra for (a) astrocytomas (n=12), (b) ependymomas (n=4), (c) medulloblastomas (n=18) and (d) controls (n=4) with 95% confidence intervals indicated by the shaded region¹².

In a follow-up study, Gly was validated as an in vivo marker for high grade childhood brain tumors. Looking at 15 tissue samples (high grade: 8 medulloblastomas, 1 anaplastic ependymoma; low grade: 5 pilocytotic astrocytomas, 1 ependymoma), the study showed a significant relation between the normalized Gly concentration measured with short echo time for in vivo MRS and the Gly levels of the corresponding samples examined with HRMAS (p<0.05). High grade tumors showed higher Gly (p<0.05) when compared to the low grade group¹³.

Results from analyses of 29 childhood brain tumors with PCA and linear discriminant analysis (LDA) agreed that different tumor types possess clear metabolomic fingerprints of significant changes for glial tumors (10 pilocytic astrocytomas and 2 ependymomas) vs. primitive neuroectodermal tumors (PNETs) (9 medulloblastomas, 7 neuroblastomas and 1 supratentorial PNET) (p<0.05)⁴¹. Glial tumors displayed higher levels of Cr and Glu and lower levels of Tau, Cho, PCho and PE when compared to PNETs, with 90% classification accuracy. Within the group of PNETs, differentiation of just medulloblastomas and neuroblastomas was feasible with 94% accuracy, as medulloblastomas show higher Cr, Gly, Glu, PCho and scy-Ino (p<0.05)⁴¹.

Metabolites and histopathological tissue properties

In addition to distinguishing among different brain tumor types and grades, HRMAS MRS has also been used to measure correlations between tissue histopathological features and metabolic properties.

Micro-heterogeneity of brain tumors is common in neuropathology, especially in high grade tumors. It complicates tumor biopsy grading and hinders correlations between metabolic changes and pathological features. Tissue HRMAS MRS can help uncover these correlations. In 2000, very soon after the invention of intact tissue HRMAS MRS, this capability was demonstrated by a study of multiple tissue specimens obtained from a 44-year old GBM patient. The specimens were first measured by ex vivo HRMAS MRS and then underwent quantitative histopathological analysis. Highly significant positive correlations between the amount of necrosis in the specimen and Lac (p<0.032, R²= 0.939) as well as lipids (p<0.02, R²= 0.961) were found. Additionally, the amount of glioma was clearly related to an increasing PCho/Cho ratio (p<0.033, R²= 0.936)⁴⁵. In addition to uncovering relationships between metabolites and brain tumor pathologies, this study, for the first time, demonstrated the critical advantage of HRMAS MRS in enabling both high resolution tissue metabolic quantifications and histopathology evaluations of the same tissue specimen. This unique capability of HRMAS MRS proved to be critical for many human studies, such as those of PCa to be reviewed later.

Apoptosis plays a major role in checking tumor growth and development. As new drugs aim to activate apoptotic pathways within the tumor cells to control their growth, a study included 41 brain tumor biopsies of grades II (n=8), III (n=3) and IV (n=30) astrocytomas revealed Tau as a potential marker of apoptosis in glioma. Tau positively and significantly correlated with apoptotic terminal deoxynucleotidyl transferase biotin-dUTP nick end labelling (TUNEL)-stained nuclei in non-necrotic (p=0.003, R=0.727), as well as in necrotic (p=0.0005, R=0.626) biopsies. Additionally, Lip/MM in the regions of 1.3 ppm (p<0.005, R=0.703) and 2.8 ppm (p<0.005, R=0.705) were positively associated with the number of apoptotic cells in non-necrotic tissue samples. These markers may help to monitor treatment response in clinic²⁵.

In addition to necrosis and apoptosis, identification of other common histopathological features is important for disease evaluation and treatment monitoring. Results from studies of 52 patients of different types of glial tumors showed metabolite differences between tissue with necrosis, high cellularity, or cancer infiltration. The best separations among the three groups were enabled by tCho/NAA and Lip/NAA ratios, with the latter being especially useful in the differentiation between necrotic and highly cellular tissue. The necrotic tissue was characterized by negligible NAA and tCho but very high quantities of lipids, whereas highly cellular tissue showed low NAA and lipids but great amounts of tCho, a marker of cell membrane turnover³³. Furthermore, 2HG was found to be positively correlated with mitotic activity (measured by the MIB-1 index), cellular density and relative tumor content within the tissue sample¹⁴.

In 2012, an ex vivo ¹H and ³¹P HRMAS MRS study of 33 brain tumor biopsies to identify metabolites, pH values, and their correlations with histopathological features was reported. Percentage of unaffected brain tissue revealed high associations with NAA, Glycerophosphorylethanolamine (GPE), GABA, Cr (all p<0.01) and Glu (p<0.05). In contrast, the percentage of necrosis was negatively correlated with Cr, GABA, Gln, Ade (all p<0.01) and positively with fatty acids (FA) (p<0.01). The metabolites Gln, PE + PCho, FA and the Glu/Gln and Cho/Cr ratios were significantly related to the percentage of tumor cells. GPE, Cr, NAA and Ade (all p<0.01) were all found to be negatively correlated with tissue pH. However, different correlations were found within only GBMs (n=19). The amount of cancerous tissue was positively linked to GSH, PC and myo-Ino levels (all p<0.05). In contrast, FA was negatively associated (p<0.05). Amount of necrosis in the GBM samples and amount of PC, Lac, Glu, GSH, and myo-Ino were inversely correlated (all p<0.05)¹⁷.

To assist readers in navigating through the large volume of HRMAS MRS studies on human brain tumors, we have summarized literature data in Table 1.

Table 1.

Selected papers on brain tumors with metabolic findings highlighted, 1998–2015. Additional abbreviations: His, histidine; Lys, lysine; Thr, threonine; Val, valine.

Year	Author	Samples investigated	Metabolites increased (in first named entity vs. second)	Metabolites decreased (in first named entity vs. second)
1998	Cheng9	Astrocytoma (low grade →high grade)	Cho/Cr	NAA
2007	Tzika38	Malignant glioma %	PCho/tCr
		Necrosis %	Lip/tCr
2008	Andronesi6	High grade tumors, metastases	Lip, FA
		Metastases (breast)		GPC
		Metastases (general)	PCho, PCho/GPC
	Davies12	Pediatric pilocytic astrocytoma	FA, isoleucine, leucine, Val, NAA, GABA, Glu	Cr, myo-Ino, Tau, ChoCC,
		Pediatric ependymoma	myo-Ino, Cr	NAA, Tau
	Erb16	High grade oligodendroglioma vs. low grade oligodendroglioma	Ala, Val	Proline, Glu, Gln, GABA, NAA
	Monleon22	High grade meningioma	PCho, PE, GSH
		Meningioma (general)		No NAA, low FA
	Opstad24	Necrosis % in GBMs	Lip/MM
2009	Davies13	High grade pediatric brain tumors	Gly
	Opstad25	Apoptotic cell density of tumor	Tau (independent of necrosis), Lip/MM (in non-necrotic tissue)
	Righi29	High grade gliomas	PCho, PCho/tCr
		Low grade gliomas		GPC, PCho/tCr
		Gliomas vs. healthy	ChoCC
	Wilson41	Pediatric glioma vs. PNETs	Cr, Gln	Tau, PE, PCho, Cho
		Pediatric medulloblastoma vs. pediatric meuroblastoma	Cr, Gln, PCho, Gly, scy-Ino
2010	Cuellar-Bana11	Medulloblastoma (general)	Tau, GPC, PCho, Cho	NAA, FA, Glu
		Pilocytic astrocytoma (general)	FA
		Ependymoma (general)	myo-Ino
	Monleon23	Atypical meningioma	Lac
	Righi30	Tumors vs. healthy	Gly
		GBM vs. low grade tumors	Gly
		GBM vs. metastases	Gly/myo-Ino
		Meningioma vs. low grade tumors	Gly/myo-Ino
	Wright43	GBM	Cr, Gly, Gln, hyp-Tau
		Astrocytoma II vs. GBM	myo-Ino	Ala
		Astrocytoma II vs. Metastasis	Gln, myo-Ino	PE
		Astro II vs. meningioma	Cr, GPC, His, myo-Ino, scy-Ino	Ala, Glu, GSH, PE
		Astro III vs. metastasis	Hyp-Tau
		Astro III vs. meningioma	GPC, myo-Ino	GSH, Lys
		GBM vs. metastasis	Cr, Gln, Gly, hyp-Tau
		GBM vs. meningioma	Asp, Cr, GPC, His, myo-Ino, NAA, scy-Ino	Ala, Glu, GSH, Isoleucine, Tau, Val
		Metastasis vs. meningioma	Cr	Ala, Glu, Gln, Lac
2011	Chen8	Increasing astrocytoma tumor grade	Lip, Lac, Ala, scy-Ino	NAA, myo-Ino, Cr
		Astrocytoma II, III vs. I	Lip, Ala	NAA, myo-Ino, Cr, PCho, GPC
		GBM (general)	Gly, Lip	NAA, PCho, GPC
		GBM vs. Astrocytoma I – III		Lac, Cr, myo-Ino
		GBM vs. astrocytoma I	Lip, Tau	Lac, Cr
		Astrocytoma III vs. I, II	Gly, Lac/Cr, myo-Ino/Cr, Gly/Cr, scy-Ino/Cr, Ala/Cr	NAA, Cr, myo-Ino
		Medulloblastoma (general)	Tau, PCho, GPC	Lac
		Medulloblastoma vs. astrocytoma II – IV	myo-Ino, Tau, Gly, PC, GPC, Asp	Lac, Cr
		Astrocytoma I – II vs. III	NAA, Cr, GPC, myo-Ino	Lac, PCho
		Astrocytoma I – II vs. GBM	Lac, Cr, Cho, GPC	Gly, PCho
		Astrocytoma III vs. GBM	Lac, Cr, Cho, GPC	myo-Ino, PCho
	Croitor33	Border tissue/cancer infiltration vs. high cellularity or necrosis	NAA	Lip/Cr
		High cellularity vs. border tissue/cancer infiltration or necrosis	ChoCC/Cr	Lip/Cr, NAA/Cr
		Necrosis vs. border tissue/cancer infiltration or high cellularity	Lip/Cr	Cho/Cr
	McKnight20	Astrocytoma II vs. III		PCho
		Ki-67	GPC
		Cell density	PCho, tCho
2012	Constantin10	Glioma II → higher grade	2HG, hyp-Tau, Cho, Lip, GSH, Ala	myo-Ino
		Secondary Glioma III vs. secondary IV	changes in hyp-Tau, GSH, Ala
		Secondary Glioma III vs. recurrent Glioma II	changes in hyp-Tau, GSH, Cho, myo-Ino, 2HG
		Likelihood of malignant transformation in recurrent low grade gliomas	Increasing 2HG levels
	Elkhaled14	Metabolites (related to malignancy) positively correlating with 2HG	Asp, GABA, Thr, hyp-Tau, Cr, PCr, Bet, Gly, Lac, GSH, PE, Glu, Gln
		Negatively correlating with 2HG		myo-Ino/tCho
	Righi31	Hemangiopericytoma vs. meningioma	myo-Ino/Glu, Glc/Glu, GSH/Glu	Cr/Glu, Gln/Glu, Ala/Glu, Gly/Glu, ChoCC/Glu, Ala
2013	Sjobakk35	Necrosis in tissue	Lip
		Brain metastases from melanoma vs. other brain metastases	GPC
	Vettukattil40	Astrocytoma II vs. GBM	GPC, myo-Ino	PCho, Gly, Lip
		Non-recurrent vs. recurrent GBM	PCho
		% necrosis	Lip, Lac
		Recurrent vs. non-recurrent GBM	GPC, myo-Ino, Cr	Gly
2014	Elkhaled15	Transformation from Glioma grades II to III or IV	2HG, tGSH	myo-Ino/tCho
		Primary glioma grade III vs. secondary	Cho, Gly, PC/GPC
		Transformation to IV	PC/GPC
		High malignancy regions	tCho	myo-Ino/tCho, NAA, Cr/PCr
2015	Kohe18	Retinoblastoma level of differentiation	Tau
		Level of necrosis	Lip	Tau, hyp-Tau, Cr, PCho, tCho
		Retrolaminar optic nerve invasion	tCho, PCho

Breast cancer

Studies of BrCa have shown HRMAS MRS as a promising tool in cancer diagnosis, treatment monitoring, and prediction of prognosis^54–73. Here, we will focus our review on four interesting aspects: metabolic differences between malignant and non-malignant breast tissues, correlations between metabolites and pathologic features of BrCa, predictions of the response to the clinical treatment and the prognosis of BrCa with metabolic information, and the relationships of metabolite, gene, and protein information.

Metabolic differences between malignant and non-malignant breast tissues

Many conventional diagnostic tools have been developed and applied in clinical practice for BrCa, including mammography, ultrasound, and dynamic contrast-enhanced MRI to provide differential diagnosis for this complex disease. Several studies of BrCa using both in vivo and ex vivo MR spectroscopy have shown that malignant lesions from the breast have increased levels of ChoCC^74,75. Studies on extracts of BrCa tissue have reported elevated levels of PC in cancerous tissue compared with non-involved tissue^76–78 and benign breast lesions⁷⁹.

In 2006, the relative intensities of GPC, PC and Cho were examined by HRMAS MRS measured from the microscopy-confirmed cancerous (n=76) and non-involved samples (n=9)⁵⁵. Consistent with previous findings, intensity ratios of PC/Cho were higher in tumor samples than in non-involved tissue (p=0.005), and GPC/PC and GPC/Cho were lower (p<0.001 and p=0.013). Classification of tumor and non-involved samples based on the relative intensities of the different ChoCC showed high sensitivity and specificity, 82% and 100%, respectively.

A study of 31 breast tissue samples (13 cancers and 18 non-cancers) again showed that in cancer cells, ChoCC, particularly PC, increased, while introducing the novel finding of increased Tau (p<0.009) and decreased Asp (p<0.002)⁵⁹. The low sensitivity and high specificity of the prediction of the cancer status were 69% and 94%, respectively.

HRMAS MRS observable metabolites discriminated between BrCa and non-involved adjacent tissue in 328 tissue samples (263 BrCa samples and 65 non-involved samples) from 228 BrCa patients. BrCa samples were shown to contain higher levels of ChoCC and Tau, in addition to ascorbate, Lac, Cr, and Gly, and lower levels of glucose (Glc) compared to non-involved tissue (Figure 4). The results emphasized ChoCC as primary biomarkers for tumor presence, given that BrCa and noninvolved tissue could be accurately discriminated with high sensitivity (91%) and specificity (93%) (p<0.001). This finding may have the potential to guide resection margins during BrCa surgery to reduce risk of local recurrence and re-surgery⁶⁶.

Figure 4.

The corresponding loading profile of first principal component (PC1), explaining 40.1% of the total variation of the data. β-Glc, β-glucose; Asc, ascorbate; Lac, lactate; Cr, Cr; Gly, glycine; Tau, Taurine; GPC, glycerophosphocholine; PCho, phosphocholine; Cho, free choline⁶⁶.

Correlation between metabolites and pathologic features of breast cancer

Histopathological (tumor size, grade, and axillary lymph node status) and immuno-histochemical [steroid hormone receptor, human epidermal growth factor receptor 2 (HER2), and cellular proliferation marker Ki-67] evaluation of BrCa specimens is the gold standard for patient treatment planning. Studies have focused on correlations between metabolic information measured from the ex vivo HRMAS MRS method and the clinical pathological results. A study of 103 samples (85 BrCa and 18 non-involved tissues) from 85 BrCa patients showed that tumors >2.0 cm were found to have significant higher levels of Gly (p=0.03) and Cho (p=0.04) than tumors ≤2.0 cm (Student’s t-test). While no significant metabolite differences were found in samples between lymph node-positive and -negative patients (Student’s t-test), a trend of increased myo-Ino in samples from node-positive patients (p=0.08) was seen⁵⁵. Separately, analysis of 36 core needle biopsy samples showed BrCa of high histological grade presenting higher PC/Cr than low histological grade BrCa (p=0.04). Higher tCho concentrations have been detected in high-grade BrCa, indicating a correlation between Cho phospholipid metabolism and tumor malignancy and aggressiveness⁶¹.

As is known in the BrCa clinic, estrogen receptor (ER) positive and progesterone receptor (PR) positive tumors present better prognosis, while HER2 overexpression tumors are associated with worse prognosis. To study the relationship between metabolites and these hormone and growth receptors, a study measured 36 ER-negative BrCa samples and found higher Cho than in ER-positive samples (p=0.03). PR-negative cancers also showed higher concentrations of Cho, as well as higher Cr and Tau, than those of PR-positive cancers (p=0.01, p=0.02, and p=0.02, respectively). Concentrations of Tau, scy-Ino, and myo-Ino of HER2-positive BrCa were significantly higher than those of the HER2-negative cancers (p=0.01, p=0.03, and p=0.01, respectively). Triple negative BrCas (TNBC) (ER-negative, PR-negative, and HER2-negative cancers) showed higher Cho concentrations and higher values of Cho/Cr and tCho/Cr than those of non-triple negative cancers. BrCa with high Ki-67 levels showed higher concentrations of tCho and PC, and higher values of PC/Cr than those of BrCa with low Ki-67 (p=0.01)⁶¹. These high concentrations of ChoCC in BrCa tissues may be a consequence of overexpression and changes in the levels of choline kinase activity in response to demands from the cell for increased phospholipid synthesis^80,81. Down-regulation of choline kinase alpha (CHKA), the gene regulating the conversion of Cho to PC, has been shown to decrease cell proliferation, and to increase the effect of chemotherapy in ovarian cancer⁸² and in BrCa⁸³, whereas CHKA overexpression increases drug resistance in BrCa cells⁸⁴. Targeting the genes or enzymes responsible for the Cho phospholipid metabolism may provide new molecular targets for treatment of TNBC⁶⁷.

In another investigation of hormone receptors and protein expression, a larger study of 106 biopsies from 73 patients showed clear separation between TNBC and triple positive BrCas (TPBC) (ER-positive, PR-positive and HER2-positive cancers) (77.7% accuracy, p=0.001) using varying levels of metabolites: TNBC is characterized with higher levels of Cho and GPC and a lower level of Cr compared to TPBC. Partial least square-discriminant analysis (PLS-DA) models enabled clear separations between ER-negative and ER-positive (72.2% accuracy, p<0.001), and PR-negative and PR-positive (67.8% accuracy, p<0.001) tumors in the testing set (Figure 5). ER-negative tumors show higher levels of Gly, Cho, and Lac compared to ER-positive tumors, suggesting enhanced glycolytic activity. PLS-DA classification and relative quantification of PR-negative and PR-positive tumors show similar metabolite profiles to ER-negative and ER-positive tumors. Validation of a PLS-DA model to separate HER2-negative and HER2-positive tumors discriminated the two with 69.1% accuracy (p<0.001). By relative quantification, HER2-negative tumors have a lower level of Gly, Gln, succinate, and Cr and higher levels of Ala, compared to HER2-positive tumors. Regardless of ER and PR status, higher levels of Gly were associated with HER2 overexpression⁶⁷.

Figure 5.

PLS-DA score and loading plots of (A) TNBC vs. TPBC, (B) ERneg vs. ER pos, and (C) HER-2 neg vs. HER-2 pos breast cancer tumors. In the score plots (left), each symbol represents one sample. The score plots show the first and second latent variables (LV), and are used for interpreting relations between samples, thus similar samples are located close to each other. In the loading plots (right), the symbols represent metabolites that are significantly important for the discrimination between the groups. Variable importance in the projection (VIP) scores is illustrated by the heat map. The majority of TNBC, ERneg and HER-2neg samples have positive score for LV1⁶⁷.

Prediction of the response to the clinical treatment and prognosis with metabolic information

As BrCa is a heterogeneous disease, major efforts aim to more precisely identify women at risk for recurrence and to predict patient responsiveness to systemic therapies. Thirty-seven tissue samples from 37 locally advanced BrCa (a subset of BrCa characterized by the most advanced breast tumors in the absence of distant metastasis, T_3–4 N_1–3M₀) patients were obtained by core needle biopsy⁶⁵. These patients were treated with anthracycline- and/or taxane-based neoadjuvant chemotherapy (NAC), and subsequently underwent surgery. The 37 samples were analyzed for a number of metabolite concentrations and choline-related compound ratios. Applying multivariate analysis, orthogonal projections to latent structure-discriminant analysis (OPLS-DA) model showed clear discrimination of three pathologic responses to NAC (sensitivity between 84.6–100%): pathologic complete response, partial response, and stable disease. Low concentrations of Tau, Cho, and GPC in the complete response group distinguished it from the others (p<0.05).

Another study confirmed that the partial response and stable disease groups could not be differentiated^63,64. Seventy pre-treatment and eighty post-treatment spectra from 85 patients, who were treated weekly with doxorubicin NAC for 16 weeks before surgery, were acquired. Although there was a significant change in tumor metabolism in response to NAC treatment for patients with partial response and stable disease, there was no difference in the metabolic response between them by PLS-DA⁶³. On the other hand, when comparing the metabolic changes between survivors (≥5 years) and non-survivors (<5 years), significant differences in the metabolic response to NAC were seen, including increases in Lac (p=0.004), and decreases Glc (p=0.002), in response to treatment in non-survivors compared to survivors. Under normoxic conditions, cancer cells, according to the Warburg effect, can reprogram their energy metabolism to largely depend on anaerobic glycolysis as their primary energy pathway, resulting in increased Lac production^85,86. Increased Lac levels may be a marker for tumor aggressiveness as high levels of Lac have been correlated with low survival rates, high incidence of distant metastasis and recurrence, and increased risk of radiation resistance in several types of cancer ^87–89. It has been hypothesized that Lac may enhance the invasiveness of tumor cells and the resulting low pH may help tumor cells evade tumor-attacking immune cells⁹⁰.

From a larger open-label multicenter study where patients were randomly allocated to receive NAC treatment with either anthracycline (epirubicin) or taxane (paclitaxel) monotherapy, samples were analyzed from a subcohort of breast cancer patients (n=89)⁶⁴. In this study, Gly appears to be decreased in survivors (≥5 years), and the difference in Gly relative intensities before and after treatment was significant (p=0.047)⁶⁴. The observed Gly decrease in survivors may reflect altered glycolysis and/or reduced Cho levels associated with reduced tumor aggressiveness⁶⁴.

HRMAS MRS was applied to intact tissue samples from BrCa patients with clinically defined good (n=13) and poor (n=16) prognoses⁵⁷. Good prognosis was defined as no detectable cancer spreading to axillary lymph nodes, tumor smaller than 2 cm, and positive for ER and PR hormone receptors, with poor prognosis as the opposite. Absolute quantitative measurements of nine metabolites were analyzed by peak area ratios and PCA. A trend toward lower concentrations of Gly in patients with a good prognosis (1.1 mmol/g) compared to patients with a poor prognosis (1.9mmol/g) (p=0.067) was found. Tissue concentration of Glc was found to negatively correlate to proliferation index (MIB-1) (r²=−0.46, p=0.013). Most importantly, high levels of Tau, GPC, and Cr combined with low levels of Gly and PCho in BrCa tissue characterized patients who remained healthy five years after surgery. Combined peak area ratios or PCA of metabolic profiles from several metabolites more strongly correlated with patients’ prognosis and long-term health status than single metabolite concentrations.

In another study, 209 tissue spectra from 156 patients were analyzed by PLS-DA, probabilistic neural networks, and Bayesian belief networks to compare the prediction capability of different multivariate analysis methods⁵⁸. A blind set (n=50) was used to test the training set (n=~150) for verification of each multivariate model. ER and PR status were successfully predicted by each method, but they were best predicted by PLS-DA with a correct classification of 44 of 50 and 39 of 50 samples, respectively. Lymph node status was best predicted by Bayesian belief network with 34 of 50 samples correctly classified, indicating a relationship between metabolic profile and lymph node status. It showed that hormone receptor-negative patients appear to have more Gly, GPC, and Cho than receptor-positive patients (p<0.05).

Relationships of metabolite, gene and protein information

To investigate the potential of evaluating molecular mechanisms at different levels, a feasibility study measuring genetic and metabolic data from the same BrCa samples was conducted for BrCa tissue samples from 46 patients. Following HRMAS MRS, microarray analysis showed no significant change in total RNA integrity (p=0.86)⁵⁶. Luminal A tumors, a subtype of BrCa that is HER2-, ER-, and PR-negative, were split into three subgroups based on hierarchical clustering of the HRMAS MR spectra. One of the subgroups, designated A2, showed significantly lower Glc and higher Ala levels than the other two groups (p<0.001), suggesting higher glycolytic activity in the A2 tumors (Figure 6). Potentially explained by increased glycolytic activity and indicative of a more aggressive tumor, Gene Ontology (GO) terms related to cell cycle and DNA repair also significantly differentiated this group from A1 tumors.

Figure 6.

Differences in metabolic and gene expression profiles of identified groups of luminal A samples. Mean spectra of the A1 (black), A2 (red) and A3 (blue) groups of luminal A tumors are plotted with the significantly different points (p<0.001) indicated with grey vertical lines. At ppm values with allocated metabolites and significant differences in one group vs. the other, the direction of the arrows indicate if the latter group has the highest (up) or lowest (down) mean value⁵⁶.

Analysis of protein expression was also combined with metabolite and gene evaluations in a study of 228 BrCa patients⁷⁰. HRMAS MRS was performed to gather tumor metabolic profiles, and hierarchical cluster analysis resulted in three significantly different metabolic clusters (Mc1, Mc2, and Mc3) that could provide treatment guidance in the clinic. Genomic and proteomic assays were also conducted to investigate correlations with the metabolomic-derived clusters: reverse phase protein array subtypes and genetic subtypes significantly differed among the three clusters, with assays finding that extracellular matrix- and collaged-associated genes were downregulated in Mc1, for example. Metabolomic evidence of an increased glycolytic rate in Mc1 also helps to classify it as the cluster likely with the worst prognosis. These approaches of combining information from several –omics levels show promise in sub-classifying BrCa or improving the understanding of breast cancer heterogeneity, potentially leading to more patient- and therapy-specific treatment⁷⁰.

Selective tissue metabolic relationships observed with intact BrCa tissues are summarized in Table 2.

Table 2.

Selected papers on breast cancer with metabolic findings highlighted, 1998–2016. Additional abbreviations: Asc, ascorbate; Asn, asparagine; Eta, ethanolamine; Fum, α-H_aa, hydrogen attached to α-carbon of amino acid, Phe, phenylalanine; pro, proline; Tyr, tyrosine.

Year	Author	Samples investigated	Metabolites increased (in first named tumor vs. second)	Metabolites decreased (in first named tumor vs. second)
1998	Cheng54	IDC II vs. II–III		Lac/Cho, PC
		IDC II vs. III		Lac/Cho, PC
2006	Sitter55	Tumor vs. non-involved	PC/Cho	GPC/PC, GPC/Cho
		IDC III vs non-cancer (tumor without cancer cells)	Cho
		Chemo-treated vs. non-cancer	Cr
		Tumor >2.0 cm vs tumor ≤2.0 cm	Gly,Cho
		Cancer cell vs. fat and fibrous connective tissue	Gly, PC
2010	Borgan56	A1 vs. A2	β-Glc, α-Glc, Lip	Ala, α-Haa, myo-Ino
		A2 vs. A3	Ala	β-Glc, α-Haa, α-Glc
		A3 vs. A1	α-Haa, myo-Ino	Lip
	Sitter57	Good prognosis vs. poor prognosis	Tau/Gly, GPC/Gly, tCho/Gly
	Giskeødegård58	ER-negative vs. ER-positive	Gly, GPC, Cho, Ala, Asc, Cr, Tau, PC, Lac
		PR-negative vs. PR-positive	Asc, Lac, Gly, GPC, PC, Cho, Cr, Ala
2011	Li59	Cancer vs. non-cancer	Tau, ChoCC (PC especially)	Asp
	Klomp60	31P MRS 7T in vivo patient 1 vs. healthy	PE, PC, GPC
		1H MRS 7T in vivo patient 1 vs. healthy		tCho
2012	Choi61	ER-negative vs. ER-positive	Cho
		PR-negative vs. PR-positive	Cho, Cr, Tau
		HER2-negative vs. HER2-negative		Tau, scy-Ino, myo-Ino
		Low histological grade vs. high histological grade		PC/Cr
		TNBC vs. TPBC		Cho, Cho/Cr, tCho/Cr
		Low Ki-67 vs. high Ki-67		PC, tCho, PC/Cr
		Good prognosis vs. poor prognosis		scy-Ino, Gly
	Giskeødegård62	ER-positive survivors vs. non-survivors		Lac
	Cao63	Pre-NAC survivors vs. post-NAC survivors	Gly, GPC, PC, Cho, tCho	β-glc, Lac, Gly
	Cao64	Pre-NAC PR vs. post-NAC PR	GPC, tCho
		Pre-NAC survivors vs. post-NAC survivors	GPC, Cho
		Survival ≥5ys vs. Survival <5ys	tCho	Lac
2013	Choi65	pCR vs. PR		PC/Cr
	Bathen66	Tumor vs. non-tumor	Asc, Lac, Cr, Gly, Tau, ChoCC	Glc
2014	Cao67	TNBC vs. TPBC	Cho, GPC, Glu	Cr, Gln
		ER-negative vs. ER-positive	Gly, Cho, Lac, Glu	Gln
		HER2-negative vs. HER2-positive		Gly, Gln, succinate, Cr
2016	Chae68	DCIS vs. DCIS with invasive cancer	GPC/PC, myo-Ino, succinate
	Yoon69	High SER vs. low SER	Asn, Cho, Fum, Glu, His, PE, Phe, Tyr, uracil, tCho
		High SUV vs. low SUV	Asn, Fum, Glu, Lac, PC, PE, uracil, tCho
		ER-positive, high SER vs. low SER	Asn, ethanol, Eta, Fum, Gln, His, Lys, Phe, Tyr, uracil
		ER-positive, high SUV vs. low SUV	Fum, Lac, uracil
		PR-negative, high SER vs. low SER	Asn, His, Phe, Tyr, Uracil
		PR-negative, high ADC vs. low ADC	Asp
		PR-negative, high SUV vs. low SUV	Asn, PE, Uracil
		PR-positive high SER vs. low SER	Fum, Lys
		PR-positive high ADC vs. low ADC		His, Phe, Tyr
		PR-positive high SUV vs. low SUV	Lac
		HER2-negative, high SER vs. low SER	Asn, Cho, Eta, Fum, His, PE, Uracil
		HER2-negative, high ADC vs. low ADC		Thr
		HER2- negative, high SUV vs. low SUV	Asn, Cho, Fum, Glu, Lac, PE, Uracil	Ile
		HER2-positive, high ADC vs. low ADC	Eta, Uracil	Lac
		HER2-positive, high SUV vs. low SUV	Betaine, Cr
		Ki67-negative, high SER vs. low SER	Fum, His, Lys, Pro, Uracil	Ala
		Ki67-negative, high ADC vs. low ADC	Leu	Tau
		Ki67-negative, high SUV vs. low SUV	Asn, Fum, Lac, Uracil
		Ki67-positive, high SER vs. low SER	Phe
	Haukaas70	Mc1(58p)	GPC, PC	Lac, Tau, Ala, Glu, Ace
		Mc2(58p)	β-glc, myo-Ino	Lac, Cr, Gly, Tau, GPC, PC, Ala, Asc, L-Tyr
		Mc3(112p)	Lac, Gly, Tau, Cr, Ala	β-Glc, GPC, PC
		Mc1 vs. Mc2	GPC, PC, L-Tyr, Asc, Cr, Ala	β-glc, myo-Ino, scy-Ino, Glu, Ace
		Mc1 vs. Mc3	GPC, PC	Lac, Tau, Ala, Glu, Ace, myo-Ino
		Mc2 vs. Mc3	β-glc, myo-Ino, Ace	Asc, Lac, L-Tyr, Gly, GPC, PC, Cr, GSH, succinate, Ala
	Park71	Center vs. periphery	Pro
	Haukaas72	1.5h after vs. before	Glc, Gly, Cho	GPC

Prostate Cancer

This chapter critically reviews publications on prostate cancer (PCa)^91–116 and concentrates on metabolic changes that allow for discrimination between benign and malignant tissues as potential biomarkers, as well as their connection to clinical parameters, including the Gleason scores (GS), pathological stage, and disease prognosis. Given the limitations of biopsies, digital rectal exams, and prostate-specific (PSA) tests, novel tools are necessary to guide disease surveillance, avoid overtreatment, and improve patient’s quality of life.

Metabolic differentiations of malignant and benign prostate tissues

HRMAS MRS studies on PCa started with attempts to differentiate samples according to their malignant or benign status, with the first measuring 199 specimens (5 GS=5, 126 GS=6, 53 GS=7, 9 GS=8, and 6 GS=9) from 82 prostatectomies. Metabolites PCho and Cho were seen to strongly contribute to the PC that significantly differentiated tissue status (p<0.005) (Figure 7b)⁹³.

Figure 7.

(A) High-resolution magic angle spinning 1H MR spectrum of intact tissue obtained from the removed prostate of a 61-year-old patient with GS 6 T2b tumors. (B) Three-dimensional plot of principal component 13 (PC13 correlates linearly with percent volume of cancer cells in tissue samples) vs. phosphocholine vs. choline. Cancerous and histologically benign (histo-benign) tissue samples from 13 patients can be visually separated in observation plane. The paired Student’s t test results (cancer vs. histo-benign from the same patients) for principal component 13, phosphocholine, and choline are 0.012, 0.004, and 0.001. Only results from these 13 patients could be evaluated with paired tests for other cancer positive samples were collected from patients from whom no histo-benign samples were analyzed. (C) The canonical plot resulting from discriminant analysis of the three variables in B presents the maximum separation between the two groups. (D) The resulting receiver operating characteristic curves indicates the accuracy of using the three variables in B to positively identify cancer samples⁹³.

Since publication of this study in 2005, ChoCC have been evaluated as discriminant metabolites for PCa in several others^{92,94,101,110–115}. PCa tissues have also been compared with other non-cancer prostate pathologies such as healthy glandular (n=20) and stromal tissues (n=20), with the former being the epithelia that are responsible for secretion and the latter, smooth muscle cells¹¹¹. When predominantly stromal tissue is compared to cancer, enrichment of ChoCC in the latter significant differentiates the two (p=0.01)¹¹². In addition, myo- and scy-Ino elevations may derive from membrane-associated phosphatidyl inositols or from inositol phosphates which serve as cellular signal transduction messengers¹⁰³. Notably, using multivariate linear regression, the myo-Ino/scy-Ino ratio was reported to correlate inversely with tumor load from studies of 108 non-malignant and 41 malignant samples (p=0.001)¹¹⁰. Furthermore, this ratio also seemed to correlate with distance to tumor foci (p=0.03). Therefore, the ratio between these isomers may be important for PCa evaluation. In addition, this distance relationship could also be seen with the (ChoCC + PC)/Cr ratio (p<0.001).

Lac and Ala levels are also enhanced in PCa tissues^{101,113–115}. Results from a study of 20 PCa samples revealed that higher concentrations of ChoCC and Lac were seen in PCa tissues than were measured in healthy glandular and stromal tissues (both p<0.01), and Cho and Ala were also significantly higher in PCa tissues when compared with stroma (both p<0.01)¹¹³. Although Lac and Ala enhancements may be caused by hypoxia which occurred during sample handling, their large ratio (Lac/Ala) seemed to be indicative of PCa, as Lac may be increased through anaerobic glycolysis¹¹⁵. These results indicated stimulation of glycolytic flux, cytosolic amino acid transformations, and protein synthesis¹⁰⁴. The altered Lac levels may result from a number of biological processes, including the Warburg effect^85,86,116, the activation of hyaluronan synthesis¹⁰⁵, the upregulation of the growth factor VEGF⁹⁷, or the hypoxia-inducible factor HIF-1α¹⁰⁶. When comparing benign samples predominantly containing glandular or stromal tissues, Lac and Ala levels did not differ significantly¹¹⁴. Together with Cho derivatives, ethanolamine (Eta)-containing compounds were reported as discriminant (15 cancer vs. 32 benign)¹¹¹. High levels of PC, GPC, PE, and GPE and low Eta were found in cancer specimens. Moreover, evaluations of metabolic ratios presented higher values for PC/GPC (sensitivity=73%, specificity=81%), PC/PE (sensitivity=73%, specificity=81%), PE/Eta (sensitivity=50%, specificity=77%), and GPE/Eta (sensitivity=64%, specificity=81%) ratios in cancer cases ¹¹¹. Since these molecules are precursors or degradation products of phospholipid membranes⁷⁴, they indicate cellular proliferation, apoptosis, and enzymatic activity in cancer status^117–122.

Another interesting class of biomolecules found in malignant specimens is the omega-6 polyunsaturated FA (n-6-PUFAs), including linoleic acid, γ-linolenic acid, dihomo-GLA, and arachidonic acid¹⁰⁹. Results from this study suggested that n-6 PUFAs did not accumulate in the adjacent non-malignant tissue. These data were in accordance with reports of potent stimulant effects of n-6 PUFAs for PCa development and aggressiveness⁹⁹, likely due to their involvement in diet, cell death, mitochondrial and cell membrane breakdown, and accelerated cell turnover in lipid-rich regions.

Concentrations of Cr do not significantly differ for different PCa pathological status¹¹³. For this reason, several studies have used Cr as a normalization factor and reported significant increases in (ChoCC+Cr)/Cit, ChoCC/Cr (both p<0.05)¹¹⁵, Tau/Cr (p=0.03), and polyamine-to-Cr (PA/Cr) ratios (p=0.013) in cancer samples. Many also report a decrease in the Cit/Cr ratio (p=0.05)¹¹² or positive correlation of (ChoCC + PC)/Cr, Cho/Cr, and scy-Ino/Cr with tumor load (0%–100%) (p<0.05)¹¹⁰. In the latter study of 149 tissue samples, 108 samples with 0% tumor load were characterized by a broad variability in stroma-to-epithelium ratio, as well as by the presence of inflammatory cells and diverse glandular morphologies. Interestingly, inter-sample difference of (GPC + PC)/Cr ratios did not vary as much for these samples with variable pathology as compared with samples of >0% tumor loads. The (GPC + PC)/Cr ratio also correlated with number of proliferating cells, measured by proliferation marker Ki67 presence (p<0.001)¹¹⁰.

Potential biomarkers in PCa

Since serum PSA levels are known to be prostate specific, but not cancer specific nor capable of predicting PCa aggressiveness, additional biomarkers may contribute to a more personalized PCa clinic⁹¹.

Results from a number of studies indicated citrate (Cit) and polyamines as potential prostate biomarkers^{92,94,112,114,115}. Bivariate linear least square analysis performed on 16 adenocarcinoma samples (9 GS=6, 6 GS=7, and 1 healthy control) showed increased levels of Cit (p=0.001) and spermine (Spm p=0.018), a polyamine, with increases in percentage volume (vol%) of normal prostatic epithelial cells in the analyzed samples⁹⁴, indicating that Cit and Spm are largely presented in histologically-benign (histo-benign) prostate epithelial glands⁹⁶. In addition, Cit and Spm were also linearly correlated (r² = 0.75, p=0.0001). The reduction of Cit is likely due to the depletion of zinc in PCa cells, since zinc can function to prevent oxidation of Cit in the Krebs’ cycle⁹⁸. Spm is also an important metabolite in the prostate for its inhibitory effect on the growth of PCa cells in vitro¹⁰⁸. Interestingly, since Cit and Spm are secreted in histologically benign (histo-benign) lumen epithelia^95,100, linear correlations were observed between concentrations of both metabolites and computer-aided image analysis-determined vol% of epithelium with lumen and without lumen. Correlations between vol% epithelium with lumen were reported for Cit (r²=0.432; p<0.02) and Spm (r²= 0.490; p<0.02), and correlations between vol% epithelium without lumen and the metabolites were r²= 0.581; p<0.01 and r²= 0.336; p<0.05, respectively. However, no correlation was found between metabolic concentrations and vol% lumen alone, suggesting the metabolites are secreted as a function of normal epithelia⁹².

Correlation of metabolomics with clinical features

Metabolites measured with HRMAS MRS also have been correlated with Gleason scores, PSA levels, or risk of biochemical recurrence (BCR).

The Gleason Score (GS) is the most relevant parameter for PCa evaluation in clinic. HRMAS MRS results indicated increases in Cho, (Cho + Cr + Spm)/Cit ratio, and n-6 PUFAs, and decreases in Cit in association with high grade PCa^107,109,112. Linear regression analyses have shown that Cho/Cr, tCho/Cr, and Cit/Cr ratios significantly relate to GS (all p<0.05), and, according to Spearman’s rank tests, significant correlation coefficients link GS with GPC+PC/Cr, Cho/Cr, tCho/Cr, Cit/Cr, tCho/Cit, and (tCho+Cr)/Cit (all p<0.05) (Figure 8)¹¹⁵. Because of Cit secretion in histo-benign glandular epithelia, higher Cit/Cr ratios were measured in histo-benign glandular than in stromal tissues (p=0.0011), or in samples with <20% (p=0.017) or >20% (p=0.05) presence of PCa glands¹¹². Thus, Cit/Cr alone may not separate predominantly stromal specimens from those containing cancer or between GS 6 and GS 7 cancers. As expected, similar results were observed using Spm/Cr ratio¹¹².

Figure 8.

Ex vivo ¹H HRMAS spectra showing the Cho-to-Cit region and corresponding H&E staining patterns of excised tissue samples containing (a) predominantly prostate cancer, (b) benign predominantly stromal tissue, and (c) benign predominantly glandular tissue¹¹².

However, unlike relative concentrations, examining absolute concentrations of Cit and Spm showed significant differences in the presence of PCa, with Spm being almost absent in GS 7¹¹². Furthermore, Spm and Cit also seemed capable of separating tumors confined in both lobes from tumors invading extraprostatic tissue and from cancer confined in a single lobe, as well as recognizing benign GS 6 from benign GS 7 samples⁹³. The ChoCC/Cr ratios were reported to be significant only for samples with cancer load ≥ 20% (p≤0.05)¹¹².

A decrease in Cit concentrations in benign epithelia was correlated to faster PCa growth rates (p<0.05, n=27 from 18 patients), as low Cit/epithelia levels were symptomatic of rapid increase in PSA values (r²=0.27, p<0.034), hence higher PSA velocities. Similar results were reported when correlating Cit/epithelia with PSA densities (r²=0.31, p<0.021) and blood percent-free PSA (r²=0.53, p<0.011)⁹⁸.

HRMAS MRS has also been employed in estimation of the risk of BCR after prostatectomy. Sixteen patients with BCR were paired with 32 subjects with no BCR and matched according to age, GS, and observation period. Based on four PCs, metabolomic profiles can differentiate matched groups with and without BCR. Analysis of variance (ANOVA) and receiver-operating characteristic (ROC) provided insight that recurrence could be predicted in 71% and 78% of cases, according to the four identified PCs and all nine PCs, respectively¹⁰².

Another interesting application of HRMAS MRS was the characterization of androgen deprivation therapy effects on patients treated with Degarelix, a drug that rapidly suppresses PSA expression and castrates testosterone levels¹⁰¹. OPLS-DA of 7 untreated and 6 treated prostate samples proposed, without a validation set, that depletions in Lac and ChoCC concentrations were indicative of Degarelix treatment. As described in the previous sections, the former observation could be related to lower glycolysis, while the latter may implicate drug-induced effects on the phospholipid metabolism. These ex vivo findings support the possibility to monitor the pharmacological response of the patients by optimized in vivo ¹H MRS techniques.

Cervical Cancer and Uterine Leiomyomas

Research on the uterine organ has focused on two clinical conditions with very different underlying biochemical mechanisms.

Cervical cancer (CeCa) accounts for 10% of all female cancers in western countries¹²³. This low rate is probably the result of the implementation of screening programs over the last 50 years; in fact, in countries where such strategies are not systematically applied, CeCa is the predominant form of cancer among women¹²⁴. The majority of the literature on the discrimination between CeCa and non-malignant specimens agrees on higher levels of Cho-containing compounds (GPC and PC) (p≤0.05) as cancer indicators^125–129. While the Cho derivatives are common features in different type of cancers due to stimulated cell turnover, the correlation between oncogenic human papillomavirus and CeCa¹³⁰ also presents viral proteins E6 and E7¹³¹ as partially responsible for alterations in ChoCC.

Lipids, especially triglycerides, comprise another class of discriminants that increase with the presence of CeCa malignancy (p<0.05)^{125,127,128,132}. Current hypotheses suggest the presence of apoptotic cells¹³³ and necrosis as reasons for increases; however, the role of necrosis is under debate because of their involvement in a variety of different cancers^134–136. Results from a linear regression show that lipid peak areas fail to correlate with tumor load for samples with greater than 20% tumor, suggesting that a sizeable tumor presence may induce changes in the adjacent tissue through ‘field effects’ that cause its lipid profile to look more like cancer (p<0.05)¹³². However, no data were available to indicate the distance over which this metabolic change is detectable.

Increases of amino acid residues such as Cr (p=0.05), Tau, and Ala (p<0.001) have been observed as characteristic of tumor tissue^125,128,129. Intriguingly, Ala and Cr levels appear depleted in benign tissue from cancer patients when compared with normal tissue from non-cancer patients, suggesting metabolite depletion in benign tissue adjacent to cancer¹²⁹. Therefore, the presence of tumors alters the metabolic profile of the surrounding cells¹³⁷. Moreover, the significant presence of Tau may be an endogenous defense mechanism against tumor proliferation¹²⁵. In addition, a study of six squamous cell carcinomas, two adenocarcinomas, and eight nonmalignant hysterectomy tissues measured higher Lac and lower levels of Glc in malignant samples¹²⁵, suggestive of the Warburg effect, the preference for anaerobic lactic acid fermentation over the usual energetic pathways^{85,86,138,139}.

Results from analysis of 44 biopsies sampled before and during radiotherapy in 23 patients identified a relationship between samples with high tumor cell fraction and high tumor cell density, finding elevated levels of Lac, Cr, GPC, and PC and decreases in Glc, myo-Ino, Tau, and Cho (p<0.001) associated with each¹⁴⁰. The increases in GPC and PC and the reduced levels of Cho suggest high membrane turnover, cell proliferation, and activation of choline kinase-initiated Cho-to-PC conversion and the phosphatidylcholine pathway^74,141,142. Furthermore, the study also showed that apoptotic cell density correlated with lipid contents (r=0.95, p<0.001). Low apoptotic activity has been associated with aggressiveness and treatment resistance in many tumor types, including CeCa^143,144. Here, the positive correlation between lipid contents and highly apoptotic tissues is probably due to modifications in the composition of cytoplasmic lipid droplets¹⁴⁵. The lipid increase in cancer specimens could be enhanced when apoptosis is triggered by radiotherapy and could be an indicator of treatment success. Lac and lipids were also reported as discriminatory metabolites between adenocarcinomas and samples from patients with lymphatic invasion¹²⁵.

Cervical intraepithelial neoplasia (CIN) is a pre-invasive stage of CeCa that can function as an investigative model to explore early events in the biology of epithelial malignancies¹²⁹. Metabolomic comparisons of punch biopsy samples of 27 squamous cell carcinoma (SCC), 12 CIN, and 39 normal tissues showed SCC as characterized by longer, more saturated lipid side-chains than CIN and healthy cases, only reaching statistical significance for distinguishing SCC and healthy¹²⁶. Increased levels of Cho and Cr also characterized SCC tissue, but no significant separation was possible between CIN and normal tissues.

ChoCC (p=0.034) and lipids at 1.3 ppm (p<0.005) were helpful in separating CIN (n=14) from CeCa (n=23)¹⁴⁶. ANOVA with Bonferroni correction and t-test analyses showed that higher levels of both metabolic classes characterized tumor cases. These observations may be explained by rapid cell membrane turnover, resulting in accumulation of choline metabolites not only in the cancer cells, but also in the peritumoural tissue. Unfortunately, analysis of the stromal portion delivered no significant information. In another study investigating CIN, ChoCC measured with 73 biopsies (5 histologically normal cervix, 5 mild CIN, 40 moderate/severe CIN, and 23 invasive cancer) showed higher (p≤0.002) concentrations could distinguish cancer from high-grade CIN tissues, and could separate CIN samples of cancer-affected from CIN of non-cancer patients (p=0.0001)¹²⁹. Nevertheless, in light of the previously discussed effects on these metabolites from tissue pathological features, such as the proportion of epithelia, interpretations of the experimental results should also consider the presence of tissue pathology.

Another medical condition affecting 20 to 50% of women is uterine leiomyomas, commonly known as fibroids, which are solid, benign, and non-degenerating pelvic tumors. Intense pelvic pain, urinary disturbances, uterine bleeding, anemia, infertility, and abortion may be related outcomes¹⁴⁷. To search for a comprehensive molecular description of the condition, intact healthy (myometrium) and leiomyoma uterine tissues from 10 subjects were studied¹⁴⁸. The application of ANOVA and PC analyses revealed an enrichment of Glu and Gln and a depletion of Tau in leiomyomas (p≤0.05). In addition to these main differences between the two groups, increases in myo-Ino, Lac, leucine and isoleucine also played roles in separation of the two groups (p≤0.05).

Gastrointestinal Neoplasia

HRMAS MRS has been used as a diagnostic tool to differentiate between gastrointestinal mucosae types. In analyzing tissue biopsies from eight female and eight male healthy subjects, no measurable difference was found in the overall metabolite composition between genders. However, metabolite intensities did show dependency on the anatomical sites from where biopsies were obtained¹⁴⁹. These investigated topographical regions including antrum, duodenum, jejunum, ileum, and colon. By using a cross-validated OPLS-DA model, it was observed that the antrum, the duodenum and jejunum together, and the colon all had specific metabolic profiles, whereas the ileum overlapped with the duodenum/jejunum and the colon (Figure 9). This observation is anatomically reasonable for the ileum links the jejunum and colon. In addition, the study correlated metabolic alterations with physiological conditions in the gut. Since the stomach is exposed to low pH values and osmotic stress, the appearance of protective glycoproteins and osmolytes, such as Tau, is physiologically logical. Traces of pancreatic lipid digestion – such as lipids and Cho – are seen in duodenal spectra, and the glutathione observed may function to protect cells against radicals resulting from lipid digestion. Amino acids seen in ileal spectra might reflect energy metabolism in the gut which depends heavily on amino acids, such as Gln. The diverse microbiome of the colon results in a great deal of bacterial fermentation, which could explain the large amounts of acetate in the spectra of colonic mucosa.

Figure 9.

OPLS-DA cross-validated scores plot showing differentiation of the metabolic profiles of localized gastrointestinal compartments. Red squares, antrum; blue dots, combined duodenum and jejunum, brown dots, ileum; green rhombus, colon¹⁴⁹.

Cancers of the gastrointestinal tract account for almost 17% of all cancer cases in developed countries and about 20% in developing countries¹⁵⁰. To understand metabolic differences among cancer types and stages within a type, HRMAS MRS studies have investigated colorectal cancer (CRCa). The findings from different reports regarding metabolic changes in CRCa were similar and consistent. In agreement with the Warburg effect, higher levels of Lac and lower levels of Glc were found in CRCa tissue samples, as well as lower levels of lipids (all p<0.05)^151–153. Increased levels of cell membrane components Cho and ChoCC, a marker of cell proliferation, were found in cancer cells (all p<0.05)^151,153. Higher levels of the amino acid Gly were found in cancer as well; in one report it reached a very high level of significance with p<0.005¹⁵² and is again linked to cell proliferation as a precursor in the synthesis of glycolysis intermediates^151,152.

While many studies focus on individual metabolites, a pilot study on rectal cancer compared 14 malignant and 9 benign samples collected from 5 patients using PCA. Analysis revealed statistically significant differences in malignant-benign comparisons for a variety of spectral regions not identified with specific metabolites (all p<0.03), as well as highlighted correlations between histopathological features and spectral regions¹⁵⁴.

In addition to reporting metabolic changes between benign and cancer samples, HRMAS MRS has also been used to assess CRCa tumor (T), lymph node infiltration (N) and metastasis (M) status – the TNM-staging essential for therapy decisions and prognosis. A study analyzed 44 samples from tumor centers and 44 samples from mucosa 5 cm from the tumor margin. The results suggested different tumor stages have distinct metabolic profiles. The development from T1/2 to T3 is marked by an increase in lipids and acetate and a decrease in GPC, whereas, T4 tissue seems to be metabolically depleted when compared to T3 tumor tissues. This seemingly-unexpected decrease in metabolite activities in T4 tumors likely indicates the T3 stage experiences the most powerful growth. The study was able to correctly classify 90%, 91%, and 75% of T1/2, T3 and T4 tumors, respectively¹⁵².

Tumor growth can metabolically change the tissue environment, as explored in a study using ‘off-tumor-tissue’ (OTT) to predict the N-stage from 83 tumor and 86 OTT samples from 26 patients. OTT samples localized next to T4-staged tumors showed lower levels of Glc and formate but increased levels of certain amino acids, such as Val (p<0.05), when compared to OTT samples next to T1 – T3 tumors, likely caused by higher protein degradation in the area surrounding higher-staged tumors. In addition, OTT presented metabolic profiles that varied by N-stage. Patients with lymph node infiltration had higher levels of Leu and phenylalanine (Phe) in their OTTs (p<0.05). Higher N-stages, in general, were associated with the trend of lower triglyceride levels likely due to a higher consumption of TGs during fast growth. Interestingly, for the prediction of the N-stages, the OTT metabolic profile presented a higher predictive power (AUC=0.92) than the metabolic profile measured from tumor tissues (AUC=0.88). Furthermore, a very promising predictive result was observed when correlating five-year patient survival rates with specific metabolic changes, which indicated that a relapse in CRCa was more probable for patients with higher concentrations of Cho, isobutyrate and acetate (Figure 10)¹⁵³.

Figure 10.

Survival prediction power for CRC using HRMAS metabolite fingerprints of tumor and off-tumor samples. (A) Scatter plot of the OPLS models of tumor and off-tumor tissue samples for patients alive (■) and deceased at 5 years after surgery (red ●). (B) ROC curve for the previous model. Reprinted with permission from¹⁵³. Copyright 2013 American Chemical Society.

Being the products of bacterial fermentation, isobutyrate and acetate may indicate the possible role of the colonic microbiome in CRCa. Supporting this hypothesis, a product of bacterial cell wall degradation, isoglutamine, was found to be higher in CRCa cells than in OTT (p<0.05)¹⁵³. Besides probable microbiota differences, metabolic differences between rectal and colonic cancer have been identified. Tissue samples obtained from colonic tumors were characterized by decreased levels of Lac (p<0.005) and increased levels of arginine and acetate (p<0.005) when compared to rectal cancer biopsies¹⁵².

Gastric cancer

Both cancerous and non-cancerous gastric samples have been studied to identify metabolites characterizing gastric tissue. About 40 metabolites, with only slight and non-significant differences in relative abundance between the samples, were identified by a study examining healthy gastric mucosa¹⁵⁵. Examination of five gastric cancer tissue samples (Helicobacter pylori-positive with different tumor grades and differentiations) and 11 healthy controls found higher levels of lipids and Gly and changes in metabolite ratios such as the Cho/ChoCC and the ChoCC/Cr ratio to be indicative of gastric cancer. While the Cho/ChoCC ratio showed an increase from 1:5 in healthy tissues to 4:1 in cancerous ones, the ChoCC/Cr ratio decreased from 16:1 to 2.5:1 in cancer samples. As stated previously, larger amounts of Gly may be linked to altered glycolysis¹⁵⁶.

In addition to gastric cancer, autoimmune gastritis and Helicobacter pylori-associated gastritis were examined in 27 subjects. Twelve healthy controls were compared to five samples in each of the three above-mentioned pathological conditions. Typical findings for gastric cancer were increased levels in free Cho, Gly, Ala, lipids, and triglycerides, with the latter only found in adenocarcinoma. Two ChoCC, phosphatidylcholine and glycerophosphocholine, decreased in cancerous samples. A similar ChoCC decrease was seen for autoimmune gastritis samples, highlighting their similarity to gastric neoplasia. Tissue with Heliobacter pylori-associated gastritis more resembled the healthy controls, with phosphorylcholine as the most abundant metabolite¹⁵⁷.

Esophageal tissue

An HRMAS MRS study of 17 samples of Barrett’s metaplasia, a precursor of the esophageal adenocarcinoma, and 17 squamous epithelium tissue samples both obtained from 16 Barrett’s patients revealed metabolic alterations between the two tissue groups. In particular, a highly significant increase in the Cho/Cr ratio was seen in Barrett’s metaplasia (p<0.001)¹⁵⁸. Results from analyzing tissue from 52 control subjects and 70 samples (32 tumor and 38 proximally histologically benign mucosa) from 35 esophageal cancer patients indicated metabolite changes in the various esophageal tissues. Applying PLS-DA on spectra of all samples, 5% of the profile showed a consistent change from healthy controls to benign mucosa to cancerous tissues from cancer patients. Metabolites that contributed to this relationship include phosphocholine and Glu and were suggested to be putative biomarkers. This study further created a model that could predict the three different types of tissue with high accuracy (p<0.01)¹⁵⁹.

Lung cancer

Lung cancer (LuCa) is the leading cause of cancer death in the United States with 155,870 estimated deaths in 2017 alone⁵. Findings from HRMAS MRS studies on LuCa overlapped for many metabolites, with specific findings in Table 4. The main aim of these studies was to reveal metabolic differences between healthy and cancerous tissue and to use these results as a tool for disease discrimination. Results showed that Lac, ChoCC and Tau were consistently elevated in cancer, whereas Glc, acetate, myo-/scy-Ino and Gly were found to be decreased^160–162. As previously discussed, the decrease in Glc and increase in Lac is likely linked to the Warburg effect, and the higher levels of ChoCC are markers for cell-proliferation. Differences in osmolytes such as myo-/scy-Ino might be linked to osmolytic shifts in tumor cells. Statistical analysis revealed significant changes in some metabolites when comparing tissue from centers of tumors with adjacent or parenchyma tissue from 17 patients [8 SCC, 7 adenocarcinoma (AC), 1 small cell lung carcinoma, and 1 malign mesothelioma]. Asp, Lac and PC/GPC showed increases from margin to tumor center (all p<0.05), whereas Glu and Val decreased progressively from margin to tumor center (all p<0.05). In addition, a positive correlation was seen between Lac and lipid levels (r=0.701, p<0.0001), but a negative correlation between Glc and Lac was observed (r=−0.439, p<0.005). Further, PCA and cross-validated OPLS-DA approaches were used to create a differentiation tool based on HRMAS MRS data. PCA could discriminate between the different sample origins – center, parenchyma, adjacent – and OPLS-DA differentiated stage I/II from III/IV and adenocarcinoma from squamous cell carcinoma (Figure 11)¹⁶².

Table 4.

Selected papers on lung cancer with metabolic findings highlighted. Samples in italics denote agreement between two or more studies.

Study	Sample size	p - values	Metabolites elevated in lung cancer	Metabolites decreased in lung cancer
Duarte161	24 patients, 26 tumor + 25 adjacent		Lac, UDP, GPC, Tau, GSH, PC	Gly, scy-ino, Glc, Lys, PE, methionine, acetate, myo-ino,
Rocha160	12 samples, 10 used for metabolomics, 2 for assignment and stuff	0.0717	Cho (+5%)	Acetate (−55%)
		0.706	Cr (+29%)	Ala (−5%)
		0.847
		0.175	GPC (+103%)
		0.00454	Lactate (178%)	Glc(−68%)
		0.00868	PCho (119%)
		6.09 × 10−7	Tau (34%)
		0.0463
		0.171
		0.286		Gly (−14%)
		0.0574		Inosine/adenosine (−35%)
		0.0504		Myo-inositol (−26%)
		0.101		Phe (−36%)
		0.612		Scy-ino (−26%)
		0.112		Tyr (−34%)
Chen162	17 patients, 3 samples from each		Lac, Ala, Asp, Tau, GPC, PCho, Lipids	Glc
Jordan163	14 tissue-serum pairs from 14 LuCa patients + 5 healthy serum controls	ANOVA p < 0.0005	Significant differentiation between cases and controls

Figure 11.

The OPLS-DA score plots and the regression coefficient plots of the HRMAS ¹H NMR spectra of lung tissue samples for various stages (n=50) and subtypes (n=44). (A) The OPLS-DA score plot and (B): the regression coefficient plot produced using only the group stage I–II and the group stage III–IV; (C) the OPLS-DA score plot and (D); the regression coefficient plot produced using only the class SCC and the class AC. Tissue at stage I–II (filled diamonds), tissue at stage III–IV (open diamonds), SCC tissue (dark triangles), AC tissue (open triangles)¹⁶².

Lung tissue HRMAS MRS results obtained from 24 tumor (11 adenocarcinoma, 4 epidermoid, 1 myofibroblastic, 3 carcinoid, 3 sarcomatoid, 2 large cell carcinoma) and 24 control samples classified 95% of the LuCa and 100% of the control samples correctly from a PLS-DA model. LuCa was characterized by elevated levels of PC, GPC, lipids, Cr, Tau, and Lac, whereas non-LuCa tissues typically showed higher amounts of acetate, methionine, and Glu. In addition to identifying LuCa from controls, metabolic differences among different tumor types, including epidermoid, carcinoid, and myofibroblastic, were observed. While the myofibroblastic type resembled the non-LuCa tissue, statistical analysis with PLS-DA allowed for proposed but unvalidated discriminations between adenocarcinoma and carcinoid or epidermoid subtypes¹⁶¹.

Besides measuring LuCa tissues, HRMAS MRS has been applied to search for new serum markers of LuCa by analyzing 14 tissue-serum pairs from LuCa patients (5 AC and 9 SCC) and 7 healthy serum controls. Although scientifically classified as liquids, using HRMAS on serum reduced the influence of serum macromolecules that hamper resolution in traditional liquid NMR. The study attempted to define metabolomic profiles of serum samples both independent of and linked to tissue metabolomic profiles of known pathology quantities. When metabolomic profiles generated from tissue spectra were applied to serum spectra, a significant differentiating power among the three serum groups (ANOVA p<0.0001) was observed. Metabolomic profiles generated solely from the serum spectra could also differentiate between SCCs and ACs (p<0.0001) and between control and AC samples (p<0.0005). This study emphasizes the possibility of applying HRMAS MRS not only on biopsies but also on serum to search for biomarkers and early diagnosis of LuCa as a screening tool¹⁶³.

Kidney cancer

Kidneys face many metabolic challenges, as they are constantly exposed to osmotic stress and high oxygen needs. There exist two kinds of renal tissues, the cortex and the medulla, with different physiologic functions, and thus different metabolite activities. Kidney cancer tissues were among the few human specimens first analyzed by HRMAS MRS after the invention of the method in the late 1990s¹⁶⁴. Since then, HRMAS MRS results have identified Gly and Tau to be typical in cortical renal tissues, and glycine-betaine and myo-Ino to be associated with the medulla, while various ChoCC are equally present in both tissue types. In addition, the same study of renal cell carcinoma (RCC), including three clear cell RCCs and two papillary RCCs, demonstrated decreases in the osmolytes myo-/scy-Ino that are likely due to osmotic imbalances in the tumor, and resulting increases in ChoCC¹⁶⁵. Interestingly, esterification of FA to TGs to cholesterol represents a unique finding in clear cell carcinoma, probably indicating the involvement of cholesterolesters carcinogenesis¹⁶⁵. Using PCA and LDA on HRMAS MRS data obtained from 22 tissue pairs of cancerous and adjacent kidney tissues, discrimination between adjacent benign kidney and RCC samples was achieved, and many metabolites were highly correlated with the PCs developed by statistical analysis (p<0.01)¹⁶⁶.

For early stage kidney cancer, surgeries are considered as a curative therapy. However, other therapeutic approaches have also been subject to investigation. For example, the effects of radiofrequency ablation were examined by HRMAS MRS in 10 patients (mean 26.6-month post radiofrequency ablation) to confirm the status of ‘no-evidence-of-disease’ originally assessed by CT scans and histological evaluations. Thirty-seven tissue samples were acquired as ‘benign’ and ‘malignant’ controls from a separate cohort of patients undergoing nephrectomy. Using spectra characterized as ‘ablated’, ‘malignant’, and ‘normal’, HRMAS MRS results fully corroborated the CT scans to confirm the ‘no-evidence-of-disease’ status. The study also hoped to distinguish between viable and nonviable tissue for the purposes of helping to assess ablation margins. In viable tissue, reductions of the metabolites Cr, Cho, myo-Ino and cholesterol implied the absence of cancer (Figure 12)¹⁶⁷.

Figure 12.

Representative magnetic angle spinning 1H spectra. (A) Ablated tumors. The peaks associated with lipid components are labeled by letters according to the chemical moiety within the triglyceride fragment seen above. Similar peaks in (B) malignant tumors and (C) normal tissue without ablation are unlabeled. Peaks labeled with numbers correlate to the following small molecular metabolites: (1) cholesterol (chol) C26, C27; (2) chol C21; (3) chol C19; (4) alanine (5) creatine/creatinine; (6) choline; (7) phosphorylcholine; (8) glycerophosphorylcholine, trimethylamine n-oxide, betaine, taurine; (9) UA singlet; (10) taurine; (11–12) myo-inositol; (13–16) amino acid and glucose resonances; (17) myo-inositol; (18) lactate¹⁶⁷.

Glandular cancer

A 2011 article analyzed 13 paired samples from surgically resected thyroid tissues and fine-needle aspiration biopsies. These specimens comprised papillary thyroid carcinomas (n=4), follicular adenomas (n=4), and normal thyroid samples (n=5). PCA and ANOVA analyses delivered 4 PCs that could discern different spectral features in the three pathological types. Notably, data from cytological biopsies and histological tissues showed similar trends, and revealed statistical significances among their canonical scores (p<0.0003). Results indicated separation was possible between papillary carcinomas and benign samples (follicular adenoma and normal) (p<0.017) and between neoplastic (papillary carcinoma and adenoma) and normal tissue (p<0.016)¹⁶⁸. Another study reported successful separation of thyroid neoplasia (papillary thyroid cancers, follicular lesions, and benign nodules) from controls obtained from the opposite healthy lobe¹⁶⁹. Following OPLS-DA, lower lipids and higher Lac, Cho, Tau, and several other amino acids levels were observed in the neoplasm group (p<0.0001). In particular, the lipid trend may be explained through accelerated metabolic turnover and higher membrane biosynthesis for cell propagation. Furthermore, benign and malignant samples could be differentiated by Lac due to either the Warburg effect^85,86 or hypoxia and ischemia in tumor tissues. Opposite levels of Tau and myo-/scy-Ino might relate to osmolyte imbalance (p<0.0001)¹⁷⁰. HRMAS examination of cell samples from ex vivo fine-needle aspirations produced similar results to those of the previous study, especially for amino acids components¹⁶⁹.

The adrenal glands also contain two functionally and structurally diverse tissues: adrenal cortex and medulla. The healthy portions of these compartments presented different metabolic hallmarks, with the cortex characterized by higher FA content (p<0.05) and absence of adrenaline and noradrenaline complexes¹⁷¹. Diseased, cortical adenomas (Ad) and adrenal cortical carcinomas (ACC) are difficult to differentiate with current diagnostic strategies^172–174. Ad are more commonly benign and are less likely to degenerate into severe morbidities^175,176, while ACC are less common but extremely aggressive, so separating them is crucial. Discrimination was possible, with significantly higher levels of Lac, acetate, and tChoCC characterizing ACC. Additionally, ACC displayed signs common to neoplasia: stimulated phospholipid turnover (higher ChoCC, p<0.05), anaerobic and glycolytic processes, and amino acids synthesis, probably due to the presence of necrotic tissues¹⁷¹.

Differentiating between benign and cancerous tissue is again difficult for tumorous pheochromocytomas (PCC). Recent metabolomic analysis, however, has shown the ability to separate PCCs from normal adrenal medulla based on several measured metabolites, including different noradrenaline/adrenaline ratios to indicate different activity of phenylethanolamine-N-methyltransferase. Noradrenaline was more abundant than adrenaline in PCC, while the trend was reversed in normal medulla (p<0.04). GABA and ascorbic acid levels were significantly altered when comparing PCC with normal medulla or ACC (p<0.05), and ascorbic acid seems to further positively correlate with noradrenaline, adrenaline, and their sums likely due to its involvement in catecholamine synthesis¹⁷¹. Another recent targeted metabolomic approach based on four metabolites (succinate, Glu, GSH, and adenosine triphosphate/adenosine diphosphate/adenosine monophosphate) was able to visualize the different profiles of apparently sporadic PCC, hereditary PCC, and those with germline mutations in succinate dehydrogenase subunits genes (accuracy > 80%, p=0.01)¹⁷⁷.

Liver cancer

Analyses of 16 diseased liver tissue samples from biopsies or surgeries (3 cirrhosis, 1 liver metastasis and 12 hepatitis of different stages and etiologies) and one healthy sample obtained from autopsy detected trends in metabolites and metabolic ratios to describe differences in tissue pathologies¹⁷⁸. The results showed levels of mobile FA, lipids, and Lac increasing from healthy to hepatitis to cirrhosis samples. Mobile FA presented as very characteristic of cirrhotic tissues. Metabolite ratios such as the Ala/Lys ratio decreased from 2.93 to 0.63 and 0.40 in hepatitis and cirrhosis cases, respectively. Furthermore, HRMAS MRS measurement of tissue from hepatitis-induced cirrhosis showed higher levels of glycogen¹⁷⁸.

Hepatocellular carcinoma (HCC), as the most frequent malignant liver disease, was studied with 31 tissue samples from 17 patients, including 6 low grade and 11 high grade diseases. Fourteen adjacent tissue samples from these subjects were analyzed as controls. Lac, Cho, Cr and bile acids increased, while Glc and glycogen decreased in cancerous tissue. Among these changes, mechanical obstruction due to tumor growth may account for the observed increase in bile acid. It is also noteworthy that metabolites such as Glc and Lac differed between low and high grade HCC. In addition, Cr and Gln, elevated in HCC in general, show a further increase from low to high grade tumors. Using metabolic profiles obtained from HRMAS MRS, three different groups (non-involved/low-grade, non-involved/high-grade and low-grade/high-grade) can be differentiated with R²-values of 87.7%, 91.8% and 90.0%, respectively¹⁷⁹.

Limited therapy options in many chronic liver diseases make liver transplantation the last possible therapeutic resort. To evaluate liver transplant status, biopsy samples from liver transplants were evaluated. Three biopsy samples (before removing the liver from the donor, at the end of cold storage, and after transplantation) were taken from each of the six studied livers. Among these transplanted livers, one patient developed post-graft dysfunction (PGD) with metabolic differences visible in the tissue spectrum. Unique to this sample, GPC did not decrease after transplantation, a change thought to indicate the beginning of the regeneration process in post-transplant tissue. Furthermore, the protective ‘University of Wisconsin solution’, used to perfuse the liver during cold storage, was seen in PGD spectra as a negative prognostic factor¹⁸⁰.

Liposarcoma

Aiming to confirm and potentially improve diagnosis and grading for liposarcoma, analysis by HRMAS MRS has shown statistically significant differences among 30 common adipocyte and neoplastic tissue samples, with neoplastic samples subdivided into lipoma, well-differentiated, well-differentiated-sclerosing, myxoid/round cell, dedifferentiated, and pleomorphic. ChoCC and tricyglerides (TGs) showed noticeable alterations throughout the whole sample set. The intensities of N-methyl-choline increased steadily as the degree of differentiation within the sample decreased. It reached the highest values in the pleomorphic samples and could most significantly distinguish well-differentiated from myxoid/round-cell samples (p=0.0001). Despite their similarities in adipocyte composition, benign lipomas and the well-differentiated liposarcomas could be discriminated using N-methyl-choline peak intensities (p=0.002). Samples with higher mitotic rates, such as pleomorphic and dedifferentiated, also presented increased amounts of phosphatidylcholine (PtC). Triglycerides behaved oppositely, with the highest levels in regular adipose tissue and lipoma and the lowest levels in pleomorphic and dedifferentiated tissues, likely due to their consumption for membrane synthesis¹⁸¹. Although the small number of cases (~5 each) in subgroup prevents generalization of the results, the ability of Cho and its derivates to serve as biomarkers appears promising.

Therapy effects on liposarcomas were assessed with a comparison of 5 tissues from troglitazone-treated liposarcoma patients and 5 patients treated by surgery only, with normal surrounding tissue acquired as controls. With PCho production from PtC an indicator of tissue growth rate, an increase in PtC/PCho ratio implies a decrease in tissue growth rate. Two treated tissue samples showed a PtC/PCho ratio closer to that of the untreated liposarcomas, however, suggesting a poor response to therapy¹⁸².

Applications of HRMAS MRS on studies of other non-malignant diseases

In addition to the great amount of HRMAS MRS research conducted on malignant and neoplastic diseases, several studies evaluated non-malignant diseases to investigate metabolic changes linked to pathological mechanisms.

Neurodegenerative diseases

The first proof-of-principle HRMAS MRS study on human tissues was conducted in 1997 on 15 autopsy brain tissue samples from four different brains regions (primary visual cortex, superior temporal gyrus, superior frontal gyrus and inferior temporal gyrus) of a patient with Pick disease. This study, for the first time, demonstrated that direct and quantitative measurements of individual brain metabolites could be achieved with HRMAS to produce intact tissue MRS with high resolution comparable to that obtained from aqueous solution. This discovery allowed the study to present a linear correlation between neuronal losses and reductions in tissue NAA levels in this neurodegenerative disease, a reduction also consistently found in malignant neoplasia reports. As the most affected brain region from Pick disease, the rostral inferior temporal gyrus presented an overall NAA decrease to 20% of the histopathologically healthy control regions (p<0.0001). Notably, even in the less affected regions, the superior frontal and temporal gyrus, the overall NAA was 58% (p=0.007) of the controls¹⁸³.

Further quantitative HRMAS MRS investigation of NAA and neuronal densities was conducted with autopsy brain samples of the superior temporal sulcus and frontal association cortex from Alzheimer’s disease (AD) patients (n=3) and healthy control subjects (n=4). The total NAA to Cr ratio acquired from the superior temporal sulcus was able to clearly and significantly separate AD patients from healthy controls (p<0.033), and an increase of PCho and a decrease of NAA were visible in AD samples. Most significantly, the study clearly demonstrated evidence of NAA as a marker for quantification of neuronal density by showing that total NAA concentration is linearly correlated (p<0.021)¹⁸⁴.

Neurodegenerative diseases are not limited to the elderly, for children can suffer from various neurodegenerative disorders such as the neuronal ceroid lipofuscinoses (NCL). This class of diseases refers to some of the most frequent reasons for progressive childhood encephalopathies. Here, 14 NCL (9 CLN1, 4 CLN3) and nine control samples were analyzed. As CLN3 is associated with more moderate brain atrophy and greater life expectancy (16–35 years) than CLN1 (8–14 years), the finding that CLN3 samples showed little difference from controls accords moderately with clinical outcomes. PCA identified decreased levels of NAA and increased levels inositols, Cr, and Cho compounds in CLN1 brain tissue samples, which distinguished them from CLN3 and control samples. Furthermore, decreased levels of Glu and GABA in CLN1 were observed, symbolizing the neuronal deterioration that alters inhibitory and exhibitory neurotransmitters in the brain and results in myoclonic epilepsies¹⁸⁵.

Amniotic fluid and gestational age

HRMAS experiments on amniotic fluid (AF) were concentrated on evaluations of gestational age (GA). Clinically, GA is assessed according to the time of last menstrual period and verified by crown-rump length ultrasound examination during the first 20 weeks of pregnancy. Unfortunately, evaluation at second trimester is not considered completely reliable due to the high variability in fetal size^186–188. Since GA is of pivotal importance for the choice of labor induction and delivery treatments^189,190, more accurate dating can improve clinical practice.

HRMAS MRS metabolic profiles of AF were generated for second and third gestation trimesters^191,192. Using a targeted approach and measuring metabolite concentration through the Electronic Reference To access In vivo Concentrations (ERETIC) method^10,193, HRMAS MRS results presented increases in creatinine and decreases in Glc, Cit, pyruvate, Cr, Ala, Glu, Lys, and valine with advancing GA (p<0.001). A linear GA prediction model (R² = 0.926)^191,192 was proposed that incorporated Ala, Cr, Glc, and Val. Interestingly, a better estimate was obtained for third trimester pregnancies, possibly due to the larger sample size (71 vs. 24 samples) or superior performance of the model in late gestation. Since AF collection through amniocentesis is an invasive procedure, the proposed model could have greater impact with the development of high resolution in vivo MRS techniques.

Cartilage tissue

A combination of metabolic changes and mechanical burden can lead to degenerative disc disease. HRMAS MRS examined intervertebral disc degeneration (IVDD), including eight Thompson grade 1, six Thompson grade 3, and six Thompson grade 5 punch biopsies from 17 discs of human cadaveric lumbar spines. Results showed statistically significant decreases changes (p<0.05) in two metabolite ratios according to increasing Thompson grade: N-acetyl/Cho, Cho/carbohydrate. Breakdown of chondroitin sulfate, an extracellular matrix component, leads to the observed reduction of N-acetyl and increased carbohydrate levels, evidencing disc degeneration¹⁹⁴. HRMAS MRS results were further corroborated by proteoglycan assay, where amino acids derived from collagen (hyroxyproline, Gly) were found to be increased in anular disc tissue, indicating breakdown of the extracellular matrix component collagen. As occurs with matrix breakdown, a smaller N-acetyl peak enables the peaks of collagen-derived amino acids to be visualized since they occur at the same resonance. MRS results obtained from nuclear disc tissue were also verified by proteoglycan assay; nucleus disc tissue has less collagen, so instead a decrease in proteoglycans (PG) signified degeneration¹⁹⁵.

In addition to linking collagen and proteoglycan breakdown with higher levels of hydroxyproline or proline, changes in carboxylic acids and alcohols were also identified as markers of degeneration. More specifically, scy-Ino and 2-propanol were shown to be the characteristic metabolites of degenerated discs from a study of 81 annulus samples (control n=21, IVDD n=60) from subjects undergoing surgery for degenerative disc disease or trauma, with the latter being used as controls. The measured metabolites were found to be age-dependent. In IVDD, the concentration of Lac and 2-propanol increased with the patient age, whereas the intensity of the n-acetyl-peak decreased. Furthermore, when considering energy metabolism in degenerative discs, reductions in Cr, an important ATP carrier, were found. The appearance of 2-propanol might be a consequence of a lack of energy metabolites that forces cells to find new ways to generate ATP, such as via ketone-based methods. Catabolic pathways such as ketone metabolism or anaerobic glycolysis lead to more acidic pH in tissue. Acidity slows down collagen synthesis and enhances the activity of matrix metalloproteinases which break down collagen¹⁹⁶. It is noteworthy that statistically significant data has also been published on discogenic back pain. When compared to tissue samples obtained from scoliosis patients, patients with pain presented lower PG/collagen (p<0.05) and PG/Lac (p<0.05) but elevated Lac/collagen ratios, suggesting a lack of energy supply that again led to higher Lac and lower pH values¹⁹⁷.

Skin

To study psoriasis arthritis, a rheumatic disease that affects the joints and the skin, skin biopsies were taken from three regions of the 10 patients: uninvolved skin, untreated psoriatic lesions, and psoriatic lesions treated with topical glucocorticoids for four weeks. During the four weeks of treatment and observation prior to the biopsy, seven patients showed a good response to the steroid with improvement in erythema and little tissue infiltration, whereas the other three responded poorly. The metablolomic studies performed on the skin tissue samples showed that the untreated psoriatic lesions were characterized by lower levels of Glc and myo-Ino, and increased levels of Cho (p<0.05), likely due to enhanced cell proliferation in psoriasis. However, since glucocorticoids act as an anti-proliferative and anti-inflammatory agent, skin biopsies taken from glucocorticoid-treated psoriatic skin lesions of good response presented a metabolic profile resembled that of the uninvolved tissue samples, with elevated Glc, myo-Ino, GPC, and Gly, and decreased Cho levels (p<0.05). Unsurprisingly, in samples from patients with poor response to the steroid, metabolite improvements were less pronounced¹⁹⁸.

Cardiovascular

A metabolomics study of varicose vein tissues reported results from eight healthy controls and eight varicose vein samples. The metabolites Cr, myo-Ino and Lac were used to characterize the unhealthy tissue samples. Since Cr plays a major role as an energy supplier in muscle cells, muscle hypertrophy in the varicose samples showed larger amounts of Cr. On the other hand, the observed changes in myo-Ino and Lac were considered to be linked with the activation of NF-κB signaling pathway and hypoxic conditions in varicose veins, respectively. NF-κB signaling is involved in various pathways, most importantly in inflammatory response. Results from OPLS regression analysis were able to differentiate the two examined kinds of tissues based on metabolites (Figure 13), with an AUC of 0.92 indicating strong ROC curve accuracy¹⁹⁹.

Figure 13.

Orthogonal partial least square (OPLS) analysis separating the two classes (varicose and non-varicose veins) based on their metabolic profiles detected using MAS-¹H NMR. Probabilistic quotient normalization and unit variance settings were applied for data analysis. R²Ycum was 0.788 and Q²cum was 0.406 for orthogonal component. Q² indicates the fraction of variation predicted by the model for cross validation. One non-varicose vein samples is closer to the varicose veins group¹⁹⁹.

Ex vivo HRMAS MRS studies on TGs within cardiac tissue were used to correlate results measured from in vivo MRS. While more often overabundant than not, TGs are critical to many pathological processes of the human heart, and improving accuracy of assessment can assist in diagnostics and disease management. In this study, biopsy samples from nine heart transplant patients demonstrated a statistically significant correlation (r²=0.83) with corresponding in vivo measurements²⁰⁰.

Summary of Metabolic Observations

In reviewing studies conducted on both malignant and non-malignant diseases, common metabolic biomarker trends became apparent for only the former, due to similar processes behind most malignant neoplasia. Specifically, many studies reported that malignant tissues are characterized by enhanced growth and proliferation, increasing cellular energy demand. Lowered levels of glucose and increased lactate production are observed in tumors, which the Warburg effect suggests are due to a shift in metabolism from oxidative phosphorylation to anaerobic glycolysis. Increased levels of growth-related ChoCC were also noted. As PtC is a major cell membrane constituent, high cell and cell membrane turnover may account for increased levels of ChoCC in tumor tissue. As a tumor grows, it can encounter the problem of self-induced hypoxia, where proliferation of tumor cells outraces that of supporting blood vessels. This phenomenon is directly linked to the aforementioned Warburg effect of increased anaerobic glycolysis, leading to higher levels of glycolysis-derived Ala and Gly in addition to triglycerides. Lipid accumulation is also a metabolic shift that is linked to necrosis, which occurs when tumor growth exceeds local oxygen and nutrient supply. Necrosis can also be induced by therapeutic radiation, and observed lipid accumulation could be indicative of successful therapy.

Malignant tumors were also characterized by an osmotic imbalance indicated by changes in Tau and myo- or scy-Ino. Whereas Tau may be an endogenous defense mechanism against cancer, myo-Ino is an activator of protein kinase C, which is involved in processes of more aggressive cerebral tumors.

Lastly, the effects of ‘metabolomic fields’, whereby cancer alters the metabolic profile of adjacent, histologically benign tissue to the tumor area, could better inform biopsy procedures or surgical margin resectioning to improve diagnostic approaches in the manifold field of oncology.

Technical Remarks

HRMAS MRS methodology has undergone significant transformation since its inception. The technique has expanded from initial, proof-of-concept research into feasibility of the method to analyze intact human tissue samples of various kinds, to differentiation of various diseases, to the proposal of MRS-based metabolomics, and to validation studies which are on the verge of translating identified metabolomic biomarkers to biological and clinical application. While HRMAS MRS remains an ex vivo technique, promising correlations with in vivo biomarker studies point to the next-generation diagnostic or prognostic advances with which the methodology can assist.

Nevertheless, we urge care in application of the technology. For instance, when comparing these ex vivo metabolomic results with disease phenotypes observed clinically or even other ex vivo studies, the same protocol and experimental conditions must be applied to yield comparable results. Care must be taken when comparing specimens collected or measured under different conditions. Specifically, specimens obtained for later analysis should be frozen within 30 min of removal from a patient to avoid changes in metabolites that occur after 60 min²⁰¹. Analysis of tissue directly after excision vs. snap-frozen samples showed increased levels of 12 metabolites in frozen tissue, which might result from intracellular lysis and release of metabolites, cautioning direct comparison of fresh and frozen samples⁷². Fortunately for biobanked specimens, long-term storage prior to analysis shows no significant impact⁷².

With the increased dimensionality seen in metabolomics, multivariate analysis is increasingly becoming the method of choice for analyzing these data, for it reduces the data complexity in order to test the underlying scientific hypotheses. However, to avoid false significance arising, corrections for multiple comparisons, such as adjusted Bonferroni significance thresholds, should be used. Supervised statistical methods, such as discriminant analyses, return statistical outputs which highlight differences between samples, but these differences must be confirmed by applying the discriminant parameters to a testing or validation cohort²⁰². Otherwise, conclusions drawn are not evidence supporting a hypothesis but merely proposal of one.

Concluding Remarks and Future Directions

Results from HRMAS MRS studies of a great variety of intact human tissue samples demonstrated the utility and future potential of the technique, rendering it capable of evaluating tissue metabolomic profiles associated with physiological and pathological processes. While the method is capable of measuring intact tissue metabolites present in a given specimen, it cannot further localize metabolites onto individual pathological features. Fortunately, recent developments in mass spectrometry imaging have enabled mapping of individual metabolites onto tissue pathologies at a single cell level. The combination of tissue metabolomics measured from HRMAS MRS with metabolites localized via mass spectrometry imaging can push the boundaries for how new in vivo MRS protocols may be designed and implemented to achieve non-invasive detection, characterization, and evaluation of therapies and diseases in clinic.

Table 3.

Selected papers on prostate cancer with metabolic findings highlighted, 2001–2016.

Year	Author	Samples investigated	Metabolites elevated (in first named entity vs. second)	Metabolites decreased (in first named entity vs. second)
2001	Cheng94	GS 6 and GS 7 vs. normal		Cit
		Vol% normal epithelia		Spm
2003	Swanson112	PCa vs. healthy glandular tissue	ChoCC, Cho, Tau/Cr	Cit, PA
		PCa vs. stromal tissue	ChoCC, Tau, myo/scy-inositol
		High GS vs. lower GS	Cho	Cit, PA
2005	Cheng93	Malignant vs. histo-benign (Hb)	PCho, Cho,
		Metabolite profiles correlate with PSA
		Hb tissue can predict pTNM and perineural invasion
2006	Swanson113	PCa vs. healthy glandular and stromal tissue	PC+GPC, tCho, Lac, Ala	Cit, PA
2008	van Asten115	PCa vs. benign	tCho/Cit, Cho/Cr, (GPC+PC)/Cr, (tCho+Cr)/Cit, Lac/Ala	Cit/Cr
	Tessem114	PCa vs. benign	ChoCC, lactate, alanine	Cit, PA
	Swanson111	PCa vs. benign	PC, GPC, PE, GPE	Eta
2009	Stenman109	PCa vs. benign	PC/GPC, PC/PE, PE/Eth, GPE/Eth, n-6-PUFAs
2010	Maxeiner102	BCR vs. non-BCR	Recurrence predicted with 78% accuracy from PCA and canonical analysis
2011	Stenman110	High tumor cell fraction vs. low	(GPCho + PCho)/Cre	myo-Ino/scy-Ino, Cho/Cre and scy-Ino/Cre;
		Benign tissue adjacent to GS 7 vs. GS 6	Cho/Cr	myo-Ino/scy-Ino
2012	Dittrich98	High PSA velocity High PSA density		Cit
		High free PSA %	Cit
2013	Selnaes107	Gleason Score	(Cho+Cr+Spm)/Cit
2016	Madhu101	Degarelix-untreated PCa vs. treated and benign	Lac, Ala, tCho, PC+GPC

Abbreviations

2HG

2-hydroxyglutarate

AC

adenocarcinoma

ACC

adrenal cortical carcinoma

AD

Alzheimer’s disease

Ad

cortical adenoma

AF

amniotic fluid

Ala

alanine

ANOVA

analysis of variance

Asc

ascorbate

Asn

asparagine

Asp

aspartate

AUC

area under curve

BCR

biochemical recurrence

Bet

betaine

BrCa

breast cancer

CeCa

cervical cancer

CHKA

choline kinase alpha

Cho

choline

ChoCC

choline-containing compounds

CIN

cervical intraepithelial neoplasia

Cit

citrate

Cr

creatine

CRCa

colorectal cancer

ER

estrogen receptor

ERETIC

electronic reference to access in vivo concentrations

Eta

ethanolamine

FA

fatty acids

fMRI

functional MRI

Fum

fumarate

GA

gestational age

GABA

gamma-aminobutyric acid

GBM

glioblastoma

Glc

glucose

Gln

glutamine

Glu

glutamate

Glx

glutamine+glutamate

Gly

glycine

GO

gene ontology

GPC

glycerophosphocholine

GPE

glycerophosphorylethanolamine

GS

Gleason Score

GSH

glutathione

HCC

hepatocellular carcinoma

HER2

human epidermal growth factor receptor 2

His

histidine

Histo-benign

histologically-benign

HRMAS

high-resolution magic angle spinning

hyp-Tau

hypo-taurine

IVDD

intervertebral disc degeneration

Ki-67

cellular proliferation marker

Lac

lactate

LDA

linear discriminant analysis

Lip/MM

lipids/macromolecule

LuCa

lung cancer

Lys

lysine

Mc

metabolic cluster

MRI

magnetic resonance imaging

MRS

magnetic resonance spectroscopy

myo-Ino

myo-inositol

n-6-PUFA

n-6-polyunsaturated fatty acid

NAA

N-acetylaspartate

NAC

neoadjuvant chemotherapy

NCL

neuronal ceroid lipofuscinoses

NF-κB

nuclear factor kappa-light-chain-enhancer of activated B cells

OPLS-DA

orthogonal projections to latent structure-discriminant analysis

OTT

off-tumor tissue

PA

polyamines

PC

principal component

PCA

principal component analysis

PCa

prostate cancer

PCC

pheochromocytomas

PCho

phosphocholine

PCr

phosphocreatine

PE

phosphoethanolamine

PG

proteoglycan

PGD

post-graft dysfunction

Phe

phenylalanine

PLS-DA

partial least square-discriminant analysis

PNETs

primitive neuroectodermal tumors

ppm

parts per million

PR

progesterone receptor

Pro

proline

PtC

phosphatidylcholine

RCC

renal cell carcinoma

ROC

receiver-operating characteristic

SCC

squamous cell carcinoma

scy-Ino

scyllo-inositol

Spm

spermine

Tau

taurine

tCho

total choline (GPC+PCho+Cho)

tCr

total creatine

TGs

triglycerides

tGSH

total glutathione (reduced + glutathione disulfide)

Thr

threonine

TNBC

triple negative BrCa

TNM status

tumor, lymph node infiltration, and metastasis status

TPBC

triple positive BrCa

TUNEL

deoxynucleotidyl transferase biotin-dUTP nick end labelling

Tyr

tyrosine

Val

valine

vol%

volume percentage

α-H_aa

hydrogen bonded to α-carbon of amino acids