Introduction

Nuclear magnetic resonance, since its discovery in 1946, has been developed into a widespread, versatile scientific tool with substantial impact on the natural sciences. It is based on the interactions of radiowaves with nuclear spins in the presence of a magnetic field. The measurement of these interactions offers valuable structural and dynamic information at the molecular level, thus establishing nuclear MR as one of the most powerful techniques of analytical and structural chemistry. A few decades after its invention, nuclear MR was applied in vivo in the form of magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS) [1]. Both use the spin of hydrogen nucleus 1H, (i.e. proton) because of its high natural abundance in the human body and its high magnetic sensitivity. In MRI high spatial resolution is required, so all hydrogen nuclei are treated collectively and structural information at the molecular level is lost. On the other hand, in proton MRS, spatial resolution is partly sacrificed in order to separate hydrogen nuclei according to their chemical group (moiety). Thus MRS can detect chemical compounds, which are usually small bio-molecules called metabolites, because they participate in specific metabolic pathways. Quantification of metabolites provides noninvasively important biochemical information and characterizes certain tissue metabolic processes and their relationship to diseases.

Historically MRS followed MRI because technically it was more challenging and demanded higher and more homogeneous magnetic fields. Also, spectra provided by MRS did not easily fit the image-centered radiologic practice. Now technical advances and numerous studies have established the importance of MRS in clinical and research settings. In pediatrics, proton MRS is applied mostly to the brain, although in adults there are growing numbers of studies of other organs, such as breast and prostate [2]. The aim of this review is to introduce the basic principles and key methodological points of acquisition and interpretation of proton MRS spectra. Τhere are excellent reviews in the literature for readers interested to clinical application of MRS [3–7].

Basic principles

Every bare hydrogen nucleus (proton) inside an external magnetic field Bo has a resonant frequency ωo, also called Larmor frequency. However protons are not isolated in human tissue but belong to a complex chemical environment that contains many additional sources of magnetic field. These sources modify the bare proton’s Larmor frequency. MRS allows us to detect these modifications and get valuable conclusions about the proton’s chemical environment. Chemical shift and J-coupling are two key concepts in MRS that are related with the magnetic interaction in the proton’s environment, originating from the surrounding electrons and the nearby nuclei, respectively.

Chemical shift

The moving electrons surrounding a proton within a molecule generate a magnetic field in the opposite direction of Bo. As a result the effective magnetic field on the proton becomes lower than Bo and the proton is said to be shielded by the electrons. In general, differences in the electronic environments cause protons to experience slightly different applied magnetic fields owing to the shielding effect of the induced electronic magnetic fields (Fig. 1).

Fig. 1
figure 1

Electronic shielding in water and methane molecules. Electronic clouds (purple) in methane (left) and water molecules (right) depict 90% of the total electron density calculated using ab initio molecular simulation (Gaussian 03 W). Note the decreased electronic shielding of hydrogen atoms in the water molecule (arrows). Oxygen is more electronegative than carbon, meaning that it attracts stronger the electrons and leaves the H nuclei less shielded. Differences in electronic shielding result in differences in the magnetic field and thus differences in resonant frequencies between carbon’s and oxygen’s H atoms

Full size image

Electronic shielding causes protons with different chemical environments to yield resonance frequencies slightly shifted from the basic Larmor frequency ωo. These deviations of resonant frequency are called chemical shifts because they depend on the local chemical environment. For practical reasons chemical shift measurements are not based on the resonance position of the bare proton. Instead they are measured from the resonance frequency of the reference substance tetramethylsilane (TMS), using the formula:

$$ \updelta \left(\mathrm{ppm}\right) = 1{0}^6*\ \left(\upomega \hbox{-} {\upomega}_{\mathrm{TMS}}\right)/{\upomega}_{\mathrm{TMS}} $$

where δ is the chemical shift of a proton in a chemical group in parts per million (ppm) and ω, ωTMS are the resonance frequencies of the proton in the chemical group and the protons in tetramethylsilane, respectively (Fig. 2). The big advantage of the parts-per-million scale is that it does not depend on Bo thus producing comparable MRS measurements regardless of scanner strength.

Fig. 2
figure 2

Proton resonance frequency ranges for various chemical groups

Full size image

In MRS spectra protons are separated in peaks according to their resonance frequencies or equivalently according to their chemical shift. Although often peaks are falsely associated with specific metabolites, in fact they correspond to protons belonging to the same or very similar chemical groups, hence called equivalent protons. As a result, different metabolites sharing the same moiety contribute to the same peak. An example is the trimethylamonium group N-(CH3)3, which resonates at 3.22 ppm. Conversely, metabolites with more than one chemical group often contribute to multiple peaks. An example is the contribution of creatine to 3.02 and 3.94 ppm peaks (Fig. 3).

Fig. 3
figure 3

An adult proton brain spectrum obtained at an echo time (TE) of 30 ms and analyzed to its constituent metabolites

Full size image

J-coupling

The magnetic field experienced by a single proton can be affected in multiple ways by nearby non-equivalent nuclei. When the proton is weakly coupled with nearby nuclei through three or fewer covalent bonds then its resonant peak splits into multiple components. This effect is called J-coupling or spin-spin coupling and its strength is represented by the so-called J-coupling constant J, measured in Hz. Note that J does not depend on the main field strength, although the peak splitting can be better detected in the improved spectral resolution of higher fields. In proton MRS, line splitting is very prominent in the case of lactate, where a doublet at 1.33 ppm arises from three magnetically equivalent methyl (CH3) protons weakly coupled to the methine (CH) proton (Fig. 4).

Fig. 4
figure 4

Echo time (TE) dependence. Methyl (red) of lactate creates a double peak because of J-coupling (=7 Hz) of methyl (red) protons with the methane proton (dashed line). For TE=72 ms doublets are out-of-phase; for TE=144 ms they are inverted (at 1.5 T); for TE=288 ms they are positive and in-phase

Full size image

Methodology

There are many ways to perform and analyze MRS. None of them is considered universal or optimum. Even more important than the choice of a particular MRS protocol is the abiding attachment to it. MRS spectra are very sensitive to many experimental parameters and experience in MRS interpretation is better gained with consistent and comparable spectra obtained with the same protocol.

Water suppression

MRS can detect small metabolites in the cell in millimolar concentrations. Their signal is typically several orders of magnitude smaller than the signal of water, which dominates the spectrum (Fig. 5) because of its high concentration (30–55 mol/L) in biological tissue. The large water signal makes it difficult — though not impossible — to obtain reliable parameter estimates for the low signal resonances of metabolites [8]. Most commonly water suppression methods are used to increase detection sensitivity of weak signal from metabolites, resulting in improved resolution and reduced measurement time (Fig. 5). Usually this is accomplished with a sequence of chemical-shift-selective-saturation (CHESS) 90° radiofrequency pulses and associated spoiling gradients that are used to excite and dephase water prior to the localization sequence [9]. Other approaches use frequency-selective radiofrequency pulses with different flip angles [10], or inversion recovery of the water signal [11].

Fig. 5
figure 5

Proton spectrum with (left) and without (right) water suppression. Dotted line represents the water peak before suppression

Full size image

Localization

In MRS there are several techniques for acquisition from selected volumes within the body [12]. Generally they can be divided into single-voxel (or single-volume) and multi-voxel techniques. The most widely used single-voxel techniques are the point-resolved spectroscopy sequence (PRESS) and the stimulated echo acquisition mode (STEAM). They both employ a combination of three slice-selective excitations that intersect at the volume of interest. In the PRESS technique, a 90° excitation and two 180° refocusing pulses are used, whereas in the STEAM sequence all three pulses are 90° excitation pulses. Many studies have compared the clinical performances of PRESS and STEAM and have found differences in signal intensity, precision of volume selection, sensitivity to motion and diffusion, modulation from J-coupling, shortest attainable echo time, efficiency of water suppression and dependence on B1 gradients [12].

Chemical shift imaging (CSI) is a multi-voxel technique that collects spectra from multiple adjacent volumes of interest in a single measurement [13–15]. Chemical shift imaging is also called spectroscopic imaging or magnetic resonance spectroscopic imaging. Spatial localization is achieved with phase encoding in one (1-D CSI), two (2-D CSI) and three (3-D CSI) dimensions as in standard MRI but with no readout gradient during data collection. Volume selective 2-D CSI, called also hybrid CSI, was developed to exclude undesired tissue (e.g., high lipid concentration, areas with susceptibility effect) that contaminates the adjacent spectra. This is attained by incorporating the volume selection pulses used in PRESS or STEAM single-voxel spectroscopy into the spectroscopic imaging sequence [16, 17]. Outer volume suppression is another way to avoid undesired signals in the spectrum and uses multiple saturation bands carefully placed around the volume of interest [16, 18].

Chemical shift imaging is preferable in non-local, diffuse or extended heterogeneous diseases where the spatial distribution of the metabolites is important. Unfortunately chemical shift imaging is often associated with reduced spectral resolution, chemical-shift artifact, voxel bleeding artifact (i.e. signal contamination into the voxel-of-interest from neighboring and even distant voxels) and voxel contamination from unsuppressed lipid or water signal. On the other hand single-voxel approaches are spatially limited but more robust and reproducible because of better magnetic field homogeneity and higher signal-to-noise ratio, resulting in well-defined peaks and uniform water suppression.

Quantification

MRS quantification using mathematical processing calculates the contribution of each of the components of the MRS signal. The objective is to find all the peaks in the spectrum and calculate the areas under the peaks (not their heights!), also called peak intensities. The area under an MRS peak is proportional to the number of equivalent H nuclei contributing to this peak. The plurality and diversity of approaches for MRS quantification reflect the complexity of the subject [19–21]. Generally MRS quantification methods can be divided as follows:

  1. (1)

    Absolute or relative quantification. Absolute quantification using a calibration procedure provides metabolite concentration in standard units (e.g., millimoles per kilogram wet weight), whereas relative quantification is based on metabolite ratios, often with total creatine as the denominator.

  2. (2)

    Time or frequency domain. Time domain methods are applied on raw, unprocessed MRS data, whereas frequency domain methods are applied after the preprocessing stage, which typically includes apodization, zero filling, Fourier transform and phasing of the data.

  3. (3)

    With or without prior knowledge. Methods with prior knowledge parameterize the MRS signal using information obtained by model spectra of metabolite solutions in vitro. Other automated, so-called black box methods directly estimate signal parameters without prior knowledge and often with minimal user interaction.

Many studies describe the relative advantages of each methodology [22, 23]. In practice, all modern scanners provide spectral quantification, but usually state-of-the-art methods like LCModel and jMRUI [24] require offline data processing. The majority of MRS studies are using relative quantification because it is faster and simpler, although it is not always accurate. Also changes of metabolite ratios are difficult to interpret because they can originate from changes of the numerator, denominator or both. Conversely, absolute quantification is in principle the right approach, but it requires extra scanning time and advanced post-processing, often becoming inefficient in clinical practice. Sometimes a good alternative for relative quantification is the comparison with spectra acquired at contra-lateral voxels in non-affected healthy tissue.

Often MR spectroscopy imaging quantification results are described by metabolite maps, which depict the spatial distribution of the metabolites. Although, for demonstration purposes, these maps appear smooth, essentially they have resolution equal to the voxel dimensions of the MR spectroscopy imaging grid.

Metabolites

Among the plethora of the metabolites in the brain, MRS can detect only few that have high concentrations and relaxation times within the detection range of current technology. Utilizing an intermediate to long (144–288 ms) echo time (TE) at 1.5 T, in most cases MRS can reliably detect N-acetylaspartate (NAA), choline-containing compounds (Cho), creatine and phosphocreatine (Cr) and lactate (Lac). For short TE (<40 ms) MRS can additionally detect myo-inositol (mI), glutamate and glutamine (Glx), glucose (Gc) and some macromolecules (e.g., proteins and lipids) (Fig. 3). Table 1 summarizes the most important brain metabolites detected by proton MRS.

Table 1 Common brain metabolites detected by proton magnetic resonance spectroscopy
Full size table

N-acetylaspartate (NAA)

N-acetylaspartate appears in the MRS spectra with a prominent peak at 2.01 ppm. Smaller contributions (<20%) to the same peak come from other neurochemicals containing the acetyl-group such as N-acetylaspartylglutamate (NAAG), N-acetylglutamate and N-acetylglucosamine. Although NAA has been found in immature oligodendrocytes and astrocyte progenitor cells, it is produced and stored almost exclusively in neurons. Specifically, NAA is synthesized from aspartate and acetyl-coenzyme-A in neuronal mitochondria. Its production is associated with fundamental metabolic processes such as the neuronal lipid synthesis via the adenosine triphosphate-citric lyase pathway and the energy production via the Krebs cycle. For all these reasons, NAA is considered a marker of neuronal density, function and viability. It decreases in cases of neuron loss, such as glioma, ischemia and degenerative diseases [25–28]. During development NAA is transported out of the neurons to the oligodendrocytes. There it is degraded to aspartate, which is used for myelin and lipid synthesis. NAA is also considered a neuronal osmolyte because of its high concentration in the brain parenchyma [29].

Choline (Cho)

Choline-containing compounds peak at 3.22 ppm comprise metabolites with the trimethylamonium group N-(CH3)3. It is also referred to as choline, or total choline (tCho), or trimethylamines (TMA) peak. The main contributors of the peak are the intracellular pools of phosphocholine (PCho) and glycerophosphocholine (GPC), which play an important role in the phospholipid metabolism of cell membrane. Smaller contributions to the Cho peak originate from other trimethylamine compounds such as free choline, citidine diphosphate choline (CDP-Cho), acetylcholine and betaine. Cellular membrane turnover during cell proliferation in neoplasms and membrane disruption in demyelination result in Cho increase [28, 30].

Creatine and phosphocreatine (Cr)

The methyl groups of creatine and phosphocreatine resonate at 3.02 ppm and are practically indistinguishable for magnetic fields lower than 7 T. A smaller peak at 3.94 ppm originates from the methylene groups. Creatine is synthesized in the human body from amino acids in the kidney, liver and pancreas. An active transmembrane creatine transporter helps creatine to enter the cells. Then, using excess adenosine triphosphate it is converted to phosphocreatine, a high energy reserve that buffers cellular adenosine triphosphate/adenosine diphosphate reservoirs. Therefore the creatine peak is considered a marker of aerobic energy metabolism. The creatine peak has also been regarded as an internal standard because of the interconversion of creatine and phosphocreatine. However, in some cases (e.g., Cr deficiency syndromes, stroke, tumor and trauma) Cr has been found to decrease, whereas in others (e.g., trauma hyperosmolar cases) it has been found to increase [3, 31].

Lactate (Lac)

Lactate appears as a doublet peak at 1.33 ppm only in pathological conditions. Its shape depends on the echo time and usually it is detected at 1.5 T and at long echo times of 144 ms (at 1.5 T) or 288 ms (at 1.5 T and 3 T) where the adjacent lipid peaks are absent. Lactate is the end product of anaerobic glycolysis and therefore it is considered a marker of hypoxia or mitochondria dysfunction [32, 33]. It increases in stroke, high-grade tumors, abscesses, mitochondrial disorders, inflammatory response and macrophage infiltration [34–36]. In neonates, especially at 1.5 T, lactate can be confused with propylene glycol, a solvent in pentobarbital [37].

Myo-inositol (mI)

Myo-inositol is a pentose sugar visible at short echo time MRS spectra at 3.56 ppm. Other smaller contributors to the same peak are the inositol monophosphate and diphosphate and phosphatidyl inositol. Myo-inositol is an important cell osmotic regulator and also participates in the phosphoinositide signal transduction pathway. It is considered to be a glial marker because it is only present in glia. It increases in inflammatory disease from glial proliferation or increased glial cell size. It is also elevated in trauma [38] and some degenerative diseases (e.g., gliosis, astrocytosis, Alzheimer disease, dementia) [39]. It may be decreased in stroke, tumor, infection and low-grade malignancies [5].

Glutamate and glutamine (Glx)

At 2.2–2.6 and 3.6–3.8 ppm there are overlapping peaks of glutamate (Glu), glutamine (Gln), gamma-aminobutyric acid (GABA), aspartate and glucose. They are labeled as a composite peak Glx because it is difficult to separate them in MRS spectra, at least at 1.5 T. Glutamate is the most abundant neurotransmitter in the brain, located in neurons and to a lesser extent in glial cells. Glutamine is the primary derivative of glutamate and is present in astrocytes. They both participate in the detoxification and regulation of neurotransmitters. GABA is an important inhibitory neurotransmitter derived from glutamate. Glx peaks have been found to increase in hypoxic–ischemic injury, hepatic encephalopathies, schizophrenia and epilepsy [4].

Lipids (Lip)

In pathological conditions and with short echo times mobile lipid and macromolecule resonances become visible. Major peaks appear at 1.3 ppm for (CH 2 )n and at 0.9 ppm for CH 3 , and smaller resonances appear at 2.05 ppm (CH2-CH 2 -CH=), 2.2 ppm (−OOC-CH 2-CH2-) and 2.8 ppm (=CHCH 2-CH=). They have been found in high-grade tumors, abscesses, acute inflammation and acute stroke. They are usually produced from cell membrane breakdown or from bacterial metabolism [40].

Other metabolites

Many metabolites (e.g., taurine, NAAG, scyllo inositol, ethanolamine, histidine, glycogen), usually with small contribution to the normal MRS spectrum, are practically undetectable. Some of them increase under disease conditions and thus can be quantified. Some examples include acetone, phenylalanine, succinate, pyruvate, alanine, glycine and threonine [41–44].

Age and location dependence

A special difficulty on the interpretation of pediatric MRS is the variation of metabolite concentration during brain development (Fig. 6). Several studies have provided normative curves describing the dependence between age and levels of metabolites [45–49]. They have found an increase in NAA and Cr and a decrease in Cho and mI (Fig. 7). Rapid changes occur within the first year of life and then the rate of change declines, with final adult values being reached until about 20 years of age. Similar curves should be used with caution because they depend on their acquisition protocol. Ideally, each site should produce its own normative curves.

Fig. 6
figure 6

Normal age-related changes of proton spectra (TE=25 ms) from a single voxel located in the basal ganglia. TE echo time

Full size image
Fig. 7
figure 7

Normative magnetic resonance spectroscopy curves according to Kreis et al. [46]. GA gestational age

Full size image

Brain MRS spectra also vary with anatomical location [50, 51] (Fig. 8) (Table 2). Generally Cho and NAA signals are higher in white matter than in gray matter. Conversely Cr, mI and Glx are less concentrated in white matter than in gray matter. The highest Cr and Cho values are found in the cerebellum.

Fig. 8
figure 8

Single-voxel proton magnetic resonance spectroscopy from a 24-year-old man. Top is from gray matter, middle is from white matter, and bottom is from thalamus, demonstrating location dependence of magnetic resonance spectroscopy

Full size image
Table 2 Age and location dependence of metabolite concentration (mean ± standard deviation in mM) obtained from single voxel STEAM pulse sequences (TE/TR =6,000/20 ms). The data are adapted from Pouwels et al. [50]
Full size table

In pathological cases the choice of the MRS location is also very important. In chemical shift imaging, phase shifting in k-space can be used to shift voxels in one or more dimensions to better enclose the region of interest [15]. In single-voxel techniques the placement of the voxel and its size depend on the clinical question. For example in tumors or in suspicious lesions the voxel must be placed in the center and its size has to be small enough to prevent contamination with nearby healthy tissue. Of course smaller voxels require an increased number of total scans and therefore longer acquisition times, but in cases of small lesions they give reliable and useful information. Another example is diffuse brain injury, where the single voxel must be placed to gray matter, far from blood and lesions to rule out global hypoxic injury [52–54]. Gray matter is more susceptible to the effects of glutamate excitotoxicity, the main cause of neuronal death after ischemia [55], because of the ambudance of postsynaptic glutamate receptors and increased synaptic activity. On the other hand if there is no suspicion of hypoxic injury, white matter is preferred to rule out diffuse axonal damage. Similarly in cases of severe hypoxic ischemia in both preterm and term neonates the deep gray matter nuclei are an ideal location for single-voxel MRS because of the known vulnerability of these regions to hypoxic–ischemic injury [56, 57].

Echo time dependence

Spectra acquired with short echo times (20–35 ms) have high signal-to-noise and they allow contributions from protons with short T2 such as the myo-inositol, the Glx compounds and the lipids. However baseline distortions, originating from poorly resolved broad resonances from short T2 and J-coupled protons, perplex the interpretation and the quantification of the spectra. On the other hand, on intermediate (135–144 ms) or long (270–288 ms) echo times, spectra become less informative but simpler and easier to quantify, with better defined baseline (Fig. 9).

Fig. 9
figure 9

Short echo time (TE=35 ms, left) and long echo time (TE=144 ms, right) normal magnetic resonance spectroscopy proton spectrum of the white matter of a 36-year-old man. Note the absence of Glx, mI and Lip/MM in the long echo time spectrum. Glx glutamate and glutamine, Lip lipids, mI myo-inositol, MM macromolecules

Full size image

Spin echoes in MRS sequences are usually created with 180° refocusing pulses, which invert the spin dephasing caused by magnetic field inhomogeneous effects. These pulses do not refocus the effects of J-coupling, which results in a TE-dependent phase modulation of the spectra. The lactate doublet, especially, appears inverted at odd multiples of TE=1/J=144 ms (Fig. 4). This phenomenon can be used to highlight the inverted lactate doublet against nearby non-inverted lipid peaks.

Field dependence

Common clinical MR systems have magnetic field strengths of 0.2–3 T. Generally, higher magnetic fields increase spin polarization and provide better signal-to-noise ratio and increased sensitivity, creating the possibility of obtaining spectra from smaller volumes or faster acquisitions. Additionally they improve spectral resolution (i.e. chemical shift dispersion), which in turn allows for more reliable quantitation of MRS spectra and better water suppression. However, field inhomogeneities and enhanced susceptibility effects, both prominent in high fields, may produce chemical shift misregistration, distort baseline and reduce resolution and sensitivity. Moreover in higher fields relaxation times are longer and the saturation effects are larger. Overall, moving from 1.5 to 3 T can probably improve a signal-to-noise ratio at least 50%. Particular care must be given in the cases of high fields (3 T and above) at the anomalous J-modulation effect, which results in lactate signal reduction and underestimated quantification [58].

Future of MRS

Compared to the progress of nuclear magnetic resonance in chemistry, MRS is still at its infancy. The reason lies in the challenges and hurdles imposed by the in vivo measurements. Low signal-to-noise ratio, field inhomogeneity, water suppression and localization pulses are basic MRS factors that hinder the full in vivo exploitation of the wealth of techniques in nuclear magnetic resonance. Studies have shown that improvements at 3 T and 4 T are below the theoretical predictions and increased resolution was counterbalanced by increased line widths [59, 60]. Nevertheless the gradual transition to higher magnetic fields and multi-frequency scanners is expected to boost MRS clinical value and research potential by allowing better spectral resolution and multidimensional techniques in acceptable scanning times. Already many studies have demonstrated the strength of 2-D MRS, but they remain far from the clinical field, mainly because of their long acquisition times [61, 62].

Another promising field of progress is the hyperpolarized MRS using carbon-13 (13C) or nitrogen-15 (15N) [63, 64]. 13C hyperpolarized pyruvate achieved with dynamic nuclear polarization along with strong fields and low temperatures has allowed unprecedented real-time visualization of the biochemical pathways of normal and abnormal metabolism [64, 65].

MRS has an important but complementary diagnostic role and supplements the clinical examination and conventional MRI by providing specific metabolite/biochemical information. Although MRS can be considered a method of molecular imaging, its low sensitivity dictates a good knowledge of the principles and limitations of this technique when it is applied and interpreted in clinical practice.