Magnetic Resonance Spectroscopy Studies of Human Metabolism

The versatility of MRS provides a hugely flexible technique capable of probing a broad range of metabolic applications across a variety of tissues. Despite its many advantages, particularly for studies of human metabolism where noninvasive techniques are at a premium, effective implementation of MRS in vivo requires consideration of its methodological limitations and how they influence its application in specific organs. In many instances, these issues can be overcome by the methodological setup and a judicious choice of the pulse sequence used to excite and detect the MR signal.

As an analytical technique, MR is relatively insensitive compared with other modalities, and tissue metabolite concentrations in the millimolar range are required, generally, to be detectable in vivo. The low natural abundance of some MR-visible isotopes (e.g., ¹³C is only 1.1% of total carbon) may further compromise detection, resulting in long scan times. Maximum sensitivity can be achieved by using an appropriate radiofrequency (RF) coil for signal detection, optimized for the nucleus of interest. Resonances of highly abundant compounds will dominate the spectrum and may obscure or distort peaks of interest. Water is an overwhelming signal in the ¹H spectrum of most tissues; ¹H and ¹³C spectra of muscle and liver feature prominent lipid peaks. Because numerous peaks are observed in ¹H and ¹³C spectra, spurious signals of other species, or unwanted signal infiltration from surrounding tissues (e.g., subcutaneous fat, the scalp), may overlap peaks of interest and impair the ability to resolve or quantify a particular metabolite.

Pulse sequences are available that suppress specific peaks (e.g., water suppression) (Fig. 2), that excite only resonances of interest, or that confine the MR signal to a specific volume (voxel), creating a localized spectrum (Fig. 3). Spectra can also be simplified by removing (“decoupling”) the dipolar coupling interaction between adjacent nuclei (e.g., ¹³C bonded to ¹H), which causes peak splitting, or by spectral “editing” to selectively enhance the signal from the metabolite(s) of interest. Typically, several of these options are combined to generate the best quality in vivo spectrum.

The MR properties of a compound or tissue may also modulate the extent to which a metabolite is visible or resolvable in vivo. Large proteins and macromolecules give rise to broad, nonspecific signals (visible as baseline distortions in brain ¹H spectra), and metabolite signals from liver and adipose tissue are significantly broader than those from muscle and brain because of the shorter T₂ relaxation times in these tissues. Spectral resolution is also compromised by effects that perturb the homogeneity of the local magnetic field within the region of interest; this is particularly apparent at air-tissue-bone interfaces (e.g., the sinuses), which exhibit differences in magnetic susceptibility. Resolution can be enhanced by correcting local field inhomogeneities by applying additional external magnetic-field gradients across the region of interest (“shimming”).

All MR techniques are susceptible to artifacts induced by motion, whether a result of inadvertent movement of the subject, respiration, the cardiac cycle, or skeletal muscle contractions. These include the infiltration of unwanted regions into the volume of interest (e.g., the gall bladder during liver MRS), compromised voxel selection, and the blurring of spectral resolution. Breath hold is a common strategy during torso and abdominal MRI, but the longer acquisition times associated with MRS favor respiratory gating, typically at end expiration when chest movement is minimal. MRS of the heart is typically accomplished with combined heart-rate and respiratory gating to compensate for both cardiac and respiratory motion. Although gating increases the experimental duration, the improvement in spectral quality is significant. Blood also is a major contaminant for cardiac MRS, and techniques have been developed to remove or correct for signal infiltration or artifacts induced by blood flow.

A detailed analysis of MRS pulse sequences, hardware, and techniques is beyond the scope of this review, but thorough descriptions can be found in the excellent book by de Graaf (1).

In vivo studies of human metabolism can be loosely grouped into four categories: 1) static estimates of metabolite content (e.g., to examine a particular subject population or the effects of an intervention); 2) dynamic monitoring of metabolites to investigate the effects of an intervention (e.g., exercise, insulin stimulation) in real time; 3) calculation of unidirectional metabolic fluxes of equilibrium exchange reactions by magnetization transfer techniques; and 4) estimation of metabolic reaction rates using MR-visible tracer kinetics. The following section will give a broad overview of the application of these techniques to investigate human metabolism in vivo.

The most widely implemented in vivo MRS technique is the use of localized ¹H spectroscopy to estimate lipid content. In muscle, intramyocellular lipid (IMCL) and extramyocellular lipid (EMCL) peaks can be resolved (Fig. 3) as a result of slight differences in chemical shift between each compartment, an effect dependent on the relative orientation of the muscle fibers with respect to the field of the MR scanner (2). As a result of variations in composition, lipid content typically is expressed relative to the water or creatine peak used as internal standards. Extending initial findings in muscle biopsy studies, IMCL content was observed to be inversely correlated with insulin sensitivity in lean, healthy individuals (3) as well as in obese and type 2 diabetic subjects (4–6), suggesting a role for the accumulation of lipid in the mechanism of insulin resistance (7,8). A notable exception occurs in endurance-trained athletes, who have increased IMCL content despite remaining highly insulin sensitive (6), although this apparent paradox may be explained by enhanced conversion of detrimental lipid intermediates (e.g., diacyglycerol, ceramides) to triglyceride (TG) (9,10).

Analogous ¹H-MRS techniques can measure intrahepatic lipid (IHL), which has similarly been correlated with liver insulin resistance (8,11,12). Interventions that decrease IHL, including drug treatment (12–14) and diet and exercise (15–17), significantly improve hepatic insulin sensitivity. These methods are the first MRS techniques to achieve widespread implementation because data acquisition and analysis are relatively rapid (<1 h to measure both IMCL and IHL) and can be performed on clinical MR scanners. Our group now includes ¹H-MRS lipid measurements as a screening tool to help identify insulin-resistant individuals, and the techniques are also being used as a biomarker to assess the efficacy of lifestyle intervention in the TuLIP (Tuebingen Lifestyle Intervention Programme) trial (15,18). MRS estimates of IHL also have become the gold standard for the diagnosis of nonalcoholic fatty liver disease and steatohepatitis and have revealed interesting ethnic differences in the prevalence of fatty liver (19,20).

The presence of myocardial TG also can be observed using cardiac and respiratory-gated localized ¹H-MRS (21). Their accumulation has been correlated with BMI (22) and impaired glucose tolerance, occurring prior to the onset of overt type 2 diabetes or left ventricular systolic dysfunction (23).

The bioenergetic state of a tissue can be characterized by using ³¹P-MRS to estimate the concentrations of high-energy phosphate (HEP) compounds or, more typically, their ratios. Unlike biochemical assays, only metabolically active (unbound) molecules are measured. Gross differences in muscle fiber–type composition can be distinguished on the basis of HEP ratios (24), and abnormalities also are observed in dystrophic muscle (25). The phosphocreatine (PCr)-to-ATP ratio is decreased in many cardiac disorders, including dilated cardiomyopathy, left ventricular hypertrophy, coronary artery disease, and heart failure, as well as infarcted heart (26–28). Bioenergetic deficits may predict clinical severity (27,29,30) and are accompanied by decreased creatine content, determined by ¹H-MRS (31), which also characterizes nonviable postischemic myocardium (32). Despite normal left ventricular morphology, mass, and systolic function, diabetic heart exhibits a decreased PCr-to-ATP ratio, which is associated with underlying diastolic dysfunction (33). Hepatic HEPs also can be measured, and decreases in ATP and inorganic phosphate (P_i) have been observed in nonalcoholic steatohepatitis (34) and type 2 diabetes (35), suggesting that impairments in liver energetics may be associated with the development of steatosis. Phosphoesters are detected in hepatic ³¹P spectra, and alterations in phosphomonester and phosphodiester content have been observed in diffuse liver disease (36). Increases in the phosphomonester-to-phosphodiester ratio may be correlated with the severity of disease (37,38), enabling ³¹P-MRS to offer an alternative to histopathology for assessing progression from steatosis to fibrosis to cirrhosis and hepatitis.

Edited and nonedited ¹H-MRS techniques also have been used to observe the effects of cerebral disease on brain metabolite content. Lactate content is elevated in infarcted regions following a stroke, whereas the neurologic marker N-acetyl aspartate (NAA) is decreased (39). Decreases in NAA have been detected in epileptogenic regions of the brain, with colocalized derangements in HEP content, suggesting that mitochondrial loss and a bioenergetic deficit may be a contributing factor (40). The inhibitory neurotransmitter, γ-aminobutyric acid (GABA), also is decreased in epileptic subjects and can be restored by Vigabatrin treatment (41), accompanied by a reduction in seizure occurrence (42).

The classic dynamic MRS experiment is to monitor the bioenergetic response of ³¹P metabolites to a sustained muscle contraction or exercise. To maintain the intracellular ATP concentration during this increased energy demand, PCr is depleted, P_i accumulates, and intracellular pH may also be modulated as a result of anaerobic glycolysis (Fig. 4). Although the intracellular concentration of free (unbound) ADP is too low to be observed directly with ³¹P-MRS, it can be calculated indirectly, assuming equilibrium at creatine kinase (CK). Upon cessation of exercise, each metabolite recovers to baseline levels. With relatively rapid (<~8 s) time resolution and certain assumptions, a range of bioenergetic fluxes can be determined from the metabolite changes, including the CK flux and glycolytic and oxidative contributions to ATP production. A functional estimate of mitochondrial oxidative capacity can be calculated from the kinetics of PCr recovery. A detailed analysis of these methods is beyond the scope of this article, but several comprehensive reviews are available in the literature (43).

Several studies have investigated the energetic response of muscle to different workloads and conditions to examine the regulation of muscle energy production during exercise. Significantly, these studies have illustrated that PCr recovery kinetics are dependent not only on the end-exercise concentrations of PCr and ADP but also on pH (44). Therefore, to estimate mitochondrial capacity accurately, particularly when assessing populations with impaired muscle bioenergetics, these factors must be considered and controlled. MRS studies have demonstrated that muscle oxidative capacity is increased by endurance training, consistent with mitochondrial biogenesis and enhanced aerobic performance, and that the metabolic perturbations associated with exercise are attenuated (45). The metabolic economy of muscle contraction during exercise has been found to decrease as intensity increases (46). During a ramped exercise protocol, elderly subjects exhibit less muscle acidification and are more resistant to fatigue (47), potentially as a result of less reliance on glycolytic ATP production and an undiminished mitochondrial oxidative capacity (48). Bioenergetic insufficiencies in muscle also have been observed in patients with cardiovascular disease (49,50), chronic fatigue (51), and type 2 diabetes (52). Dynamic ³¹P-MRS may emerge as a clinical test for the early diagnosis of ischemic heart disease because abnormal cardiac energetics are observed in response to a handgrip exercise stress test (53,54), an effect that is apparent in the absence of angiographically significant coronary stenoses (55).

Myocellular oxygenation can be assessed using ¹H-MRS to monitor the appearance of deoxymyoglobin (oxymyoglobin is effectively undetectable) when a mismatch occurs between oxygen demand and supply (56) (e.g., during exercise or ischemia). The appearance of deoxymyoglobin in exercising muscle is approximately proportional to the workload at low to intermediate intensities and then reaches a plateau (57). Intracellular oxygenation is dependent on contraction frequency as well as intensity (58), and perfusion insufficiencies can be detected in peripheral artery disease (59). Lactate production in exercising muscle, attributed to anaerobic glycolysis, can be indirectly estimated from rates of intracellular acidification. Direct measurements of lactate using edited ¹H-MRS are possible in muscle (60,61), although they remain to be fully validated. In contrast, transient increases in brain lactate during photic stimulation can be directly observed using ¹H-MRS because of the lack of overlapping lipid peaks (62).

Intramuscular glycogen stores can be detected with ¹³C-MRS and are depleted by longer-duration exercise (63); repletion occurs via insulin-dependent and independent mechanisms (63) and is influenced by diet (64). IMCL utilization also may occur, dependent on exercise duration and workload (65). The transition between the fed and fasting states has also been investigated using MRS techniques. During short-term fasting, liver glycogen is depleted (66), followed at longer durations (>72 h) by significant accumulation of IMCL (67) and IHL (68), which is likely driven by the increase in circulating plasma fatty acid concentrations. Postprandially, glycogen is replenished in muscle and liver (69) by both direct and indirect pathways of glycogen synthesis (70,71). Using a combination of MRS and non-MRS techniques, the contribution of hepatic gluconeogenesis to endogenous glucose production was estimated to be ~50%, even during the earliest time points of a fast, which is significantly higher than expected (66,72). In type 2 diabetes, elevated rates of fasting endogenous glucose production were found to be caused by increased gluconeogenesis rather than net hepatic glycogenolysis (73).

The unidirectional fluxes that contribute to metabolic exchange reactions can be examined using magnetization-transfer (MT) techniques. The underlying concept of these methods is that if the MR signal of a metabolite is perturbed, the transfer of this effect to its exchange partner can be detected, with the magnitude of the MT effect proportional to the kinetics of exchange (Fig. 5). Unidirectional metabolic fluxes can then be calculated using classical reaction kinetics. Bioenergetic rates can therefore be investigated using ³¹P-MT spectroscopy. Decreased CK (PCr → ATP) flux is observed in congestive heart failure (74) and left ventricular hypertrophy (75), suggesting that myocardial ATP generation is impaired in the pathophysiology of human heart disease. Rates of P_i → ATP flux have been estimated in resting muscle and were observed to be decreased in insulin-resistant elderly individuals (76) and insulin-resistant offspring of type 2 diabetic patients (77), indicating a bioenergetic deficit in these individuals. In contrast, in endurance-trained muscle, characterized by enhanced mitochondrial content and oxidative capacity, this basal flux is unchanged (78). Feasibility studies in human visual cortex (79) and, more recently, liver (80) also have been demonstrated. It is worth noting that high signal/noise spectra are required to accurately estimate the MT effect and that if the observed metabolite is involved in multiple exchange reactions more complex kinetic models and multiple-site MT methods are required (e.g., to measure ATP utilization). Finally, MT techniques estimate the total flux from all cellular contributions to a specific reaction, including futile cycling. These effects may lead to overestimation of the absolute P_i → ATP flux in muscle (81) and complicate the interpretation of fluxes in liver, where the contributions of nonoxidative ATP generation and glycolytic enzyme cycling may be considerable.

Rates of metabolic reactions also can be estimated by administering an MR-visible tracer and monitoring its rate of incorporation into tissue metabolites. ¹³C-MRS is ideal for such studies because the relative lack of background signal attributed to the low natural abundance of ¹³C leads to dramatic signal increases on metabolite ¹³C-enrichment (Fig. 6). Furthermore, a wide range of metabolic products can be monitored and both positional and isotopic enrichment data can be determined, which offers a significant advantage over mass spectrometry. The simplest labeling strategies monitor the accumulation of tracer into macronutrient pools. Glycogen synthesis can be readily detected in muscle and liver using labeled glucose, and insulin-stimulated muscle glycogen synthesis has been found to be decreased in type 2 diabetes (82). Similarly, postprandial liver glycogen synthesis is decreased in type 2 diabetic patients, accompanied by impaired suppression of endogenous glucose production (83). Patients with poorly controlled type 1 diabetes also have impaired postprandial hepatic glycogen synthesis (84). Ingestion of ¹³C-labeled fatty acids allows postprandial fat storage in liver and muscle to be determined, and both were found to be elevated in type 2 diabetes (85). More complex pulse-chase experiments (tracer administration followed by nonlabeled washout) enable metabolite synthesis and degradation (e.g., of liver glycogen) to be concurrently monitored. This style of study has surprisingly demonstrated that glycogen turnover occurs in liver in both fed and fasting states (86) and in muscle during steady-state exercise (87). Infusion of multiple ¹³C-labeled tracers has permitted the relative contributions of different glucose sources to liver glycogen repletion to be determined under fed and fasting conditions (70).

Sophisticated computer-modeling techniques enable more complex metabolic reaction schemes to be assessed. Rates of oxidation via the tricarboxylic acid (TCA) cycle can be estimated in muscle by using [2-¹³C]acetate as a precursor substrate and monitoring the rate of enrichment of the muscle glutamate pool, which acts as a surrogate for the intermediates of the TCA cycle (81). Decreased muscle substrate oxidation was found in two insulin-resistant cohorts of subjects (healthy older adults [77] and offspring of type 2 diabetic patients [88]), demonstrating an association between peripheral mitochondrial function and insulin sensitivity. Conversely, increased substrate oxidation rates have been observed in endurance-trained individuals (78).

Using [1-¹³C]glucose as a substrate, multiple reaction rates can be measured in brain, including the TCA cycle flux, α-ketogutarate/glutamate exchange, and a unique metabolic parameter, the rate of excitatory neurotransmitter (glutamate/glutamine) cycling between neuron and astrocyte (89,90). Glial contributions to these cerebral metabolic rates can be estimated with [2-¹³C]acetate as the metabolic precursor because of the preferential uptake by astrocytes (91). With multivoxel data acquisition plus segmentation analyses, gray- and white-matter compartmentation also can be resolved (92). Significant increases in TCA cycle oxidation rates have been observed in the visual cortex in response to visual stimulation using direct (93) and indirect (94) techniques to measure ¹³C isotopic labeling. In contrast, neurotransmitter cycling, TCA cycle flux, and glucose oxidation were found to be decreased in Alzheimer’s disease (95) and healthy aging (96). Abnormalities in brain glucose metabolism also have been observed in other mitochondrial and metabolic disorders (97). More recently, plasma lactate has emerged as a significant substrate source for brain metabolism (98).

One of the most powerful aspects of using MRS to study metabolism in vivo is that multiple independent parameters of metabolic function can be assessed across a range of tissues. This capability offers a unique ability to explore the mechanistic basis for the pathogenesis of metabolic diseases and/or syndromes. MRS studies were a key factor in elucidating the role of glucose transport in the mechanism of muscle insulin resistance in type 2 diabetes. Insulin-stimulated rates of muscle glycogen synthesis (82), measured by ¹³C-MRS, were found to be attenuated in type 2 diabetes. Crucially, ³¹P-MRS studies demonstrated that there also was a blunted increase in the intracellular concentration of glucose-6-phosphate in diabetic muscle, consistent with impaired glucose transport/phosphorylation activity, rather than a defect in glycogen synthase activity (99).

Using a multinuclear approach, a relationship between peripheral lipid deposition, muscle bioenergetics, and insulin resistance also has been demonstrated. In two cohorts of insulin-resistant subjects (76,77,88), the accumulation of IMCL (and IHL in [76]), measured by localized ¹H-MRS, was associated with decreases in substrate oxidation (TCA cycle flux) and rates of P_i → ATP flux in resting muscle, determined using dynamic ¹³C-MRS and ³¹P-saturation-transfer MRS, respectively. These studies suggest that a decline in basal mitochondrial activity, corroborated using two independent MRS biomarkers of mitochondrial metabolism, may play a role in the pathogenesis of peripheral insulin resistance and the development of type 2 diabetes. More recently, postprandial macronutrient storage in muscle and liver in response to a carbohydrate meal has been examined using ¹H-MRS to measure lipid accumulation and ¹³C-MRS to assess glycogen synthesis (100). Muscle insulin resistance alters the partitioning of carbohydrate away from muscle glycogen synthesis and toward hepatic lipogenesis.

Interleaved ³¹P/¹H-MRS techniques allow muscle HEP metabolism and myoglobin deoxygenation to be assessed simultaneously during exercise or ischemia (101) and have revealed complex interactions between vascular function, intracellular oxygenation, and mitochondrial function in elite athletes (102).

There is continuous development of novel MRS hardware, pulses, sequences, and techniques to improve the investigation of human metabolism in vivo. These occur in concert with a continual shift toward higher magnetic field strengths, which is accompanied by increased signal, wider spectral bandwidth, and enhanced spectral resolution. However, elevated power requirements at higher fields may lead to excessive localized power deposition, particularly for high-duty cycle sequences and those involving decoupling. The increased acoustic noise and space confinement of these systems may also impact subject tolerance. In this section, we highlight a few areas of development that already are generating significant interest.

Functional and metabolic heterogeneity is apparent not just between organs but also within a single tissue. Muscle HEP content is dependent on fiber type (24), and variations in IMCL distribution (103,104) and Pi → ATP flux (105) have been observed in different muscle of the calf. Differences in metabolite content and reaction rates are also apparent between regions of the brain and between gray matter, white matter, and cerebrospinal fluid (92,106). To robustly investigate metabolism in vivo, localized or spatially mapped versions of many of the techniques mentioned in this article are required. Spectroscopic imaging (SI) methods, which obtain MR spectra on a point-by-point basis across an entire region of interest, offer an attractive methodology by which this can be accomplished (Fig. 7), and certain applications already have begun to benefit from this approach (104,106–108). The recent development of multichannel detection coils augments the SI approach, concomitantly enhancing spatial coverage and decreasing scan time.

Adipose tissue has been studied in a limited manner with MRS but suffers from severe spectral broadening as a result of the short T₂ relaxation times of lipid. However, the metabolic significance of regional differences in glucose and fat metabolism between visceral and subcutaneous fat compartments, and recent studies demonstrating the existence of subclavicular brown fat in humans, is beginning to spur MRS investigation of these tissues.

MR methodology lacks a direct technique to investigate oxygen metabolism in vivo. Recently, ¹⁷O-MRS has been implemented in animal models to measure cerebral blood flow and the metabolic rate of oxygen consumption during ¹⁷O₂ inhalation. Cerebral oxygen consumption was found to increase in the visual cortex with visual stimulation and agreed well with ³¹P-MRS–determined rates of ATP synthesis (109), suggesting that this technique could successfully be translated to human research.

Despite its vast potential, ¹³C-MRS is inherently limited by low signal/noise and a broad chemical shift range. Two approaches that enhance the sensitivity of ¹³C measurements may transform its applicability to metabolic studies. Proton-observe/carbon-edited (POCE) techniques acquire ¹H-spectra but “edit” the spectrum to select protons coupled to ¹³C nuclei (Fig. 8). Using this method, ¹³C-enrichment data can be obtained with the superior localization accuracy and enhanced sensitivity of ¹H-MRS. In vivo, the application of POCE has been restricted by the relatively narrow bandwidth and spectral resolution of ¹H-MRS, but the availability of high-field systems may overcome this restriction, providing high-quality shimming routines ensure sufficient spectral resolution (110). Hyperpolarization favorably alters the Boltzmann distribution of ¹³C spins in a target molecule, increasing sensitivity by several orders of magnitude, such that hyperpolarized ¹³C-tracers can be observed with a single scan. Current applications are limited because of the requirement for specialized equipment, the limited number of metabolic substrates, and the extremely short lifetimes of hyperpolarized ¹³C. Nonetheless, in vivo studies have been performed in animal models to investigate the rapid metabolic fluxes of pyruvate (111) and lactate (112). Broader metabolic applications are yet to be developed, but this technique may potentially revolutionize in vivo ¹³C-MRS.

Chemical-exchange-saturation-transfer (CEST) is an MT method that examines generic exchange between labile metabolite protons and bulk water. Although primarily implemented as an MRI-based technique that provides exchange-based contrast, several intriguing metabolic applications are emerging. These include pH imaging (e.g., to discern tumors), glucose sensing, and imaging of glycogen distribution. CEST may also have applications as a biomarker of gene expression detecting ¹H exchange fluxes that are catalyzed by a specific enzyme, a gene delivery agent or by the RNA polymer itself. The review by Zhou and van Zijl (113) provides a comprehensive analysis of CEST techniques.

To date, the capabilities of MRS primarily have been exploited in research where it has advanced our understanding of the mechanisms behind fundamental metabolic processes and the pathogenesis of disease. More widespread availability and the translation of MRS into a practical clinical diagnostic tool has been curtailed by several factors: dedicated spectroscopy systems and specialized expertise have been required leading to high costs, scan times are relatively long, and many individuals are excluded from undergoing MR scans (e.g., those with implanted metal, pacemakers, or claustrophobia). However, the new generation of high-field (3 Tesla, and above) clinical MRI scanners can be equipped with spectroscopy capabilities, and this, combined with the advent of robust automated processing routines and rapid data-acquisition techniques, offers the scope for more widespread application of MRS techniques to human research and the potential for translation into a clinically useful methodology. In vivo MRS promises to be a key technique contributing to the investigation of human metabolism in the 21st century.

No potential conflicts of interest relevant to this article were reported.

D.E.B. and G.I.S. wrote the manuscript.