MR spectroscopy (MRS) allows the noninvasive measurement of the concentrations from selected metabolites in vivo. Till now, MR spectroscopy is applied for specific purposes in brain tumor diagnostics. The metabolic profile of a brain tumor not only characterizes tumor entity, but it may also be crucial for prognosis and for therapeutic decisions. In the last decades, it has become evident that molecular genetic markers of a brain tumor may be prognostic or even predictive for a specific therapy (Weller et al. 2009; Reifenberger et al. 2012). Therefore, therapy of brain tumors is becoming increasingly complex, and histopathological features should not be the only aspect of establishing therapeutic decisions in the future. These molecular markers influence the metabolic profile and the micro milieu of the tumor. While MRI is considered as method of choice for diagnostic imaging of brain tumors, the method of MR spectroscopy, which is based on the same physical principles as MRI and can be performed with the identical setup, provides metabolic information, thereby offering a tool for studying the metabolic profile. In vitro MRS studies of tumor specimen and many in vivo studies have already shown that MR spectroscopy is able to detect these metabolic profiles or even the oncometabolites themselves (Constantin et al. 2012). Therefore, the role of MR spectroscopy may fundamentally change in the next decades. Hitherto, MR spectroscopic studies investigated the sensitivity and diagnostic accuracy of MR spectroscopy in characterizing brain tumors and tumorlike lesions (Horská and Barker 2010). Taking into account that the accuracy is not sufficient to replace histopathological diagnosis, the value of time-consuming spectroscopic methods for differential diagnostic still remains limited. Today’s primary indications of MR spectroscopy in diagnostic settings should be: (1) guiding stereotactic biopsy in heterogeneous or large non-necrotic brain tumors, (2) avoiding surgery in asymptomatic patients with small brain tumors in eloquent brain areas or young patients with chronic partial epilepsy with “benign” aspect, and (3) monitoring residual low-grade tumors after surgery. Monitoring high-grade gliomas after standard therapy (surgery, followed by radiation and chemotherapy) may be difficult or even impossible with proton MR spectroscopy. First, recurrent high-grade gliomas often occur at the margins of resection cavity and thus in areas with preexisting damaged brain tissue from radiation, peri-surgical infarction, and macro- or microbleeds. Considering that metabolites other than lipids are only present in solid and vital tumor tissue, partial volume effects from necrotic and hemorrhagic tissue may affect the metabolite concentration obtained for the targeted voxel. Second, many malignant brain tumors are located in the temporal or frontobasal lobes. In these brain areas, proton MR spectroscopy is prone to susceptibility artifacts requiring time-consuming manual shimming or even rendering the spectra useless. Some of these disadvantages do not apply to phosphorus MR spectroscopy which measures some of the most relevant compounds involved in tumor metabolism: metabolites of membrane phospholipids, the products of oxidative phosphorylation, and the intracellular pH (see Chap. Future Methods in Tumor Imaging).

This chapter focuses on the special metabolism of glial brain tumors to elucidate the role of MR spectroscopy for a more “individualized” tumor characterization.

1 Methods

1.1 Introduction to MRS

Magnetic resonance imaging (MRI) measures the signal of water protons (1H nuclei) in the presence of magnetic field gradients, which, together with phase encoding, provides the localization. MRS detects water-soluble metabolites, thus presenting a method for in vivo monitoring of metabolic changes. At the magnetic field strength of a standard clinical MR scanner (1.5–3 T), the 1H nuclei and to a certain extent also the 31P nuclei show sufficient sensitivity to allow the in vivo detection of metabolites in small volumes (<4 ml) within reasonable acquisition time. The first in vivo 31P spectrum of mouse brain was recorded in 1978 (Chance et al. 1978). Measuring 1H spectra requires efficient suppression of the dominant water signal, which exceeds the metabolite signals by approximately 104, and is therefore technically more demanding. The first in vivo 1H spectra of rat brain were recorded in 1983 using a surface coil in a vertical bore high-resolution NMR spectrometer at 8 T (Behar et al. 1983). Human brain spectra were first obtained in 1985 by Bottomley et al. (1985).

Initially, in vivo spectra were recorded with surface coils which detect signal from the entire region in the vicinity of the coil. For application to pathological lesions, it is required to obtain spectra from a targeted region of the brain (e.g., tumor tissue). This can be achieved by pulse sequences with selective excitation of three orthogonal slabs resulting into spectra from a single cuboid volume element localized at the intersection of the slabs (single voxel spectroscopy, SVS). Two methods are available, either PRESS (Bottomley et al. 1985; Ordidge et al. 1987) or STEAM (Frahm et al. 1989), each having their advantages and disadvantages as described by Moonen et al. (1989). The potential of measuring spatially resolved spectroscopic information (i.e., obtaining a matrix of spectra as demonstrated in Figs. 1 and 2) by combining spectroscopy with gradient phase encoding (spectroscopic imaging, SI or MRSI) was first demonstrated by Brown et al. (1982) for the 31P nucleus, while 10 years later Fulham et al. (1992) showed first 1H MR spectroscopic images of patients with brain tumors. At that time many studies on human tumors had already been performed (for review, see Negendank 1992) including several 31P MR spectroscopic examinations on brain tumors applying various localization techniques. However, with the publication of the first localized water-suppressed 1H spectrum of a human brain tumor by Bruhn et al. (1989), single voxel 1H MRS had become the method of choice for supporting a noninvasive differential diagnosis in brain tumors. This can be attributed to the fact that the spectra show separate markers for the pathological and normal tissue: the increase in intensity of the choline signal is related to tumor cell proliferation (Herminghaus et al. 2002; Guillevin et al. 2008), while the decrease of the concentration of metabolite N-acetyl-aspartate marks breakdown of neuronal cells as it is shown in Fig. 1 for a patient with glioblastoma (see also Table 1).

Table 1 Important metabolites in 1H MRS and their significance in brain tumors
Full size table
Fig. 1
figure 1

MRSI parameter maps for NAA and tCho of a patient with glioblastoma. The left panel (a) shows a T2-weighted MRI with a grid overlay indicating the spatial resolution of the MRSI data. The white frame marks the area selected for spectroscopy using the PRESS excitation scheme. The glioblastoma is located left paracentral; the adjacent parasagittal cortex is slightly blurred on both sides with a mild increase in signal intensity. The color-coded maps show the regional distribution of the metabolites NAA (b, marker for intact neuronal tissue) and tCho (c, marker for proliferating cells). The tCho concentration increase is inhomogeneous showing tumor infiltration into the cortex of both hemispheres (Figure already published in Nervenartz 2014)

Full size image
Fig. 2
figure 2

Monitoring lactate. Short (30 ms, upper traces) and long (lower traces) from a patient with glioma grade IV. (a, T2-weighted MRI with a grid overlay indicating the spatial resolution of MRSI). The three panels with spectroscopic data show normal-appearing tissue (b, yellow-marked voxel in the MRI), necrotic tissue (c, central red-marked voxel in the MRI), and CSF (d, right red-marked voxel in the MRI). Lactate is visible as a doublet (two signals with 8 Hz distance) at 1.3 ppm. The signals are inverted at long TE. Note that the lipid signals at 1.2 ppm which are only visible at short TE are overlapping with lactate

Full size image

Single voxel spectroscopy relies on the accurate definition of tumor tissue from T2-weighted or CE-enhanced MRI. But these data lack information of tumor heterogeneity and potential tumor infiltration. Further, discrimination of tumor tissue may be difficult for infiltrating gliomas. In MR spectroscopic imaging (MRSI), all spectra of a selected slice are acquired simultaneously applying encoding gradients between the excitation pulse and the acquisition period. After spatial Fourier transformation, the spectroscopic image is obtained as a matrix of the dimension NX × NY, where NX and NY denote the number of phase encoding steps in each direction within the slice. For each matrix, a spectrum can be calculated representing the metabolic information for the voxel attributed to the matrix element. Signal intensities and their ratios can be visualized as a grid overlay on the anatomical image providing parameter maps for the concentrations for specific metabolites or metabolite concentrations ratios (spectroscopic image, Fig. 1). The anatomical reference image should have been recorded with identical angulation and slice offset. Such spectroscopic images provide a retrospective definition of the center and the extent of tumor tissue, while at the same time reference spectra are available from normal-appearing tissue (Fig. 2). The resolution can be as low as 0.75 × 0.75 × 1.00 cm3 at sufficient signal to noise ratio (S/N), but this requires phase encoding for the entire matrix. Consequently, acquisition of a data set with conventional MRSI techniques takes more than 15 min (see below) which might not be tolerated by many patients especially when performed in addition to the other modalities (Chaps. MR Imaging of Brain Tumors, MR Spectroscopic Imaging, MR Perfusion Imaging, and Diffusion-Weighted Methods) routinely applied in the MR examination. Modifications of the basic MRSI sequence which can reduce the data acquisition time or/and provide multi-slice data will be discussed in the next section. Details of biochemical and clinical aspects of metabolic changes will be discussed in a dedicated section.

1.2 Summary of Spectroscopic Imaging Techniques Applied in Tumor Diagnostics

For an in-plane resolution of 7.5 × 7.5 mm2, the 32 × 32 matrix shown in Figs. 1 and 2 had to be recorded at a 240 mm2 FOV. Acquisition of the entire k-space at a repetition time of 1.5 s would take 1,024 × 1.5 s or 26 min. Together with the preparation period (extensive shimming, adjustments for water suppression), the MRS examination may add another 30 min. to the conventional imaging examination. Reduction of measurement time and optimized automatic adjustments for the preparation period are therefore essential for a successful MRS protocol. The latter has been addressed in the modern scanners by the use of image-guided shimming procedures and implementations of routines for automatic adjustments of water suppression. These tools can reduce the preparation time to less than 1 min.

The rather extensive acquisition times required for the complete k-space can also be reduced. Without significant loss in spatial resolution, a 28 × 28 matrix can be recorded and extrapolated to 32 × 32 by adding zeroes before Fourier transformation, which will reduce the total acquisition time to 20 min. Selection of a circular (elliptical in case of rectangular FOV) k-space area centered around the origin will save another 25 % of acquisition time without seriously affecting the spatial resolution (Maudsley et al. 1994). The use of a rectangular FOV could also save up to 30 % (Golay et al. 2002) resulting in a total acquisition time between 10 and 15 min. Further reduction in acquisition time can be achieved with fast imaging techniques like echo planar spectroscopic imaging (EPSI) and parallel imaging method (Posse et al. 1995; Zierhut et al. 2009; Ozturk et al. 2006; Sabati et al. 2014) or multiple spin-echo spectroscopic imaging (MSESI) (Duyn and Moonen 1993). These techniques scan more than one phase encoding step for a single excitation pulse, providing the respective acceleration factors, and allow acquisition of 3D MRSI data with sufficient spatial resolution in reasonable scan time.

1.3 Partial Volume Effects Due to Low Resolution

Metabolite concentrations from lesions smaller than the grid resolution will be affected by the concentration in the surrounding tissue and changes may be masked, i.e., choline concentrations will be underestimated while NAA concentrations will be overestimated. Also, special care should be taken when nominal matrix size (i.e., the number of phase encoding steps in each direction before extrapolation by adding zeroes) is rather small (<16 × 16), since this causes significant blurring due to the poor point spread function leading to “bleeding” of signal intensity between adjacent voxels. Signal bleeding also becomes significant when the grid resolution (resolution after adding zeroes) exceeds the nominal resolution significantly; thus, digital resolution enhancement by more than a factor of 2 should be avoided. Partial volume effects should definitely be taken into account when the absolute quantification of spectroscopic data is considered.

1.4 Evaluation of Metabolite Concentrations

Spectroscopic data reflect the concentration of a subset of brain metabolites. The accuracy of the related information depends crucially on the approach used for data quantification. Generally, the spectrum is evaluated by measuring the area under the metabolite signals. This can be done either by numerical integration of metabolite peaks in phased (real) or magnitude (modulus) spectra or by using more sophisticated tools which basically perform a nonlinear fit of the entire spectrum. Depending on the tool, the fit is performed in the time domain using constraints (jMRUI (Naressi et al. 2001; Vanhamme et al. 1997), an offline tool which requires export of the data to an external workstation) or frequency domain (most processing tools which are provided by the vendor and operate on the scanner console; LCModel (Provencher 1993), offline data evaluation). All methods report signal intensities which are proportional to the respective metabolite concentration in the volume of interest (VOI). Conversion of the hardware-specific units to absolute concentrations (i.e., mMol/l) requires a set of correction factors which depend on the used pulse sequence, hardware parameters like signal amplification and coil loading, relaxation times (T1, T2) of the metabolites, as well as fractions of GM, WM, and CSF in the VOI (partial volume effects). Hardware parameters can be corrected for by using either the so-called phantom replacement method (Michaelis et al. 1993) or scaling relative to the water signal (Barker et al. 1993). The water must be recorded in a separate measurement, either as a separate MRSI data set which has to be corrected for T1 and T2 relaxations or by an imaging sequence with proton density contrast. Relaxation terms for metabolite signals from regular (healthy) tissue are available in several publications, but they may be changed in tumor tissue (Träber et al. 2004; Hattingen et al. 2007; Isobe et al. 2002). Further, the presence of contrast agents can lead to a decrease of signal intensity between 10 and 15 % (Smith et al. 2000; Sijens et al. 1997; Murphy et al. 2002). Correction for partial volume effects requires at least one more additional imaging sequence and further calculations. A rather quick method which only takes into account the CSF fraction was described by Horská et al. (2002), while analysis of GM, WM, and CSF fraction requires tissue segmentation which can be very time consuming. Therefore, a thorough data evaluation in terms of absolute concentrations should be reserved for research studies aimed at metabolic differences between different groups of patients (e.g., different tumor entities) and longitudinal studies, while for diagnostic purposes a semiquantitative approach just comparing metabolite intensities from tumor tissue and normal-appearing tissue from the contralateral side may be sufficient. Immediate information of the extent of change of metabolite concentrations or their ratios can be visualized in the MRSI metabolite map (Figs. 1 and 3). However, one should be aware of artifacts (see below).

Fig. 3
figure 3figure 3

Parameter maps for NAA (upper right) Cho (lower left) and the ratio Cho/NAA (lower right) from a glioblastoma (a) and a metastasis (b). Note the higher relative ratio Cho/NAA compared to the normal tissue in the glioblastoma (a) compared to the metastasis (b) due to a marked increase of choline signal intensity in the glioblastoma (a). The Cho/NAA ratios of the glioblastoma (a) are moderately increased outside the tumor mass, indicating tumor cell infiltration. In contrast, the Cho/NAA map of the metastasis shows clearer margins of the tumor (b)

Full size image

1.5 Artifacts in Metabolite Maps

Spectroscopic imaging data are frequently visualized as metabolite maps, i.e., for each metabolite the concentration is displayed either as a grayscale image or as a color-coded overlay on an anatomical image. While this provides the most intuitive picture of the results, special care should be taken when interpreting these maps. Local field inhomogeneities due to calcification or deposits of paramagnetic hemosiderin which occur in the vicinity of areas with former bleeding can shift and distort signals, spoiling the data analysis algorithm applied to obtain the signal intensities for the specific metabolites. Especially for voxels crucial for diagnostic decision (e.g., with highest choline), the choline hot spots or Cho/NAA signal intensities require an inspection of the entire spectrum to exclude excessive line broadening and baseline distortions which usually prohibit a reasonable signal analysis by integration or fitting routines, leading to false values for metabolite concentrations or their ratios. Intense lipid signals originating from necrotic areas as well as from fat deposits in the skull base, soft tissue, and orbit can also distort the baseline. These lipid signals can even appear in the spectra and should not be misinterpreted as tumor necrosis (Fig. 4a). An excellent description how to judge the quality of the spectra is given by Kreis (2004). Rapidly growing tumor cells typically have marked increase of glycolytic rates even if oxygen is abundant (Warburg effect (Warburg 1956), see below), and lactate is considered as a marker for increased glycolysis. Lactate in tumor tissue coincides with the lipid signal but can be easily distinguished from lipid (Kuesel et al. 1996), since only lactate shows a doublet signal (i.e., two peaks of identical intensity separated by 7.4 Hz) which will be inverted at an echo time of 135 ms (Fig. 2). At B0 field strength of 3 T, the doublet structure of lactate may be less visible due to increased line broadening at higher field strengths but signal inversion can still be exploited for discrimination of lactate from lipid.

Fig. 4
figure 4figure 4

Representative 1H spectra in gliomas of different WHO grades. Short TE (30 ms) from brain tumors depicting a low-grade (a), a grade III (b), and a heterogeneous grade IV tumor (c). Each panel with spectroscopic data shows spectra from the voxels marked in the respective MRI on the left with the upper trace referring to the tumor voxel (red) and the lower trace referring to the contralateral, normal appearing tissue voxel (yellow). Due to its more frontal position, spectra from the low-grade tumor are broadened and therefore plotted with an extended y-scale. Note: The grade III tumor shows the most enhanced tCho signal

Full size image

2 Tumor Metabolism

A major characteristic of brain tumors is the altered metabolism. In recent years it has become clear that biological modifications in tumor tissue are evident through metabolic alterations which may be of great importance in therapy resistance (Tennant et al. 2010). This chapter describes changes in metabolic pathways which are typical for tumor tissue and can be measured by MR spectroscopy. Identifying those features may be useful for the diagnosis or treatment of brain tumors.

Tables 1 and 2 show an overview of the most important 1H and 31P metabolites for brain tumors. Representative 1H and 31P spectra from gliomas with different tumor grades are shown in Figs. 4 and 5

Table 2 Metabolites measurable with in vivo 31P MR phosphorus spectroscopy in brain tumor
Full size table

.

Fig. 5
figure 5

Representative 31P spectra in gliomas of different WHO grades. 31P spectra from a low-grade glioma (a) and a heterogeneous glioma grade IV (b). Upper traces represent data of tumor tissue from the red-marked voxel in the MRI, while lower traces refer to the contralateral normal-appearing tissue from the yellow voxel in the MRI. Note: Increased GPE and PE signals in the low-grade tumor, while GPE is decreased in the glioma grade IV. Broadening of the inorganic phosphate signal in the glioma grade IV indicates increased intracellular pH in the tumor tissue

Full size image

A basic metabolic alteration in malignant cells is the phenotype which performs aerobic glycolysis even in the presence of oxygen whereas oxidative phosphorylation is suppressed (Warburg 1956). Enhanced lactic acid production through glycolysis causes extracellular acidosis. To counteract the intracellular proton accumulation, the activity of H+ extruding and buffering pathways like the Na+/H+ exchanger or the transmembrane carbonic anhydrases is upregulated (Chiche et al. 2009; McLean et al. 2000). Thus, the extracellular environment gets more acidic while the intracellular pH increases. The maintenance of an alkaline intracellular pH in tumor cells supports cellular proliferation, whereas extracellular acidosis promotes angiogenesis. Phosphorus spectroscopy is the only noninvasive method measuring both intracellular pH and the high-energy phosphate compounds ATP and PCr (Negendank 1992; Hattingen et al. 2011). Suppressed oxidative energy metabolism as a result of tumor hypoxia and repressed mitochondrial function may induce a decrease in high-energy phosphates like ATP and phosphocreatine (Papandreou et al. 2006).

Further, energy consumption is increased in neoplastic transformations to provide protein and nucleotide synthesis (Susa et al. 1989). The glycolytic pathway is linked with amino acid production. Serine as intermediate from 3-phosphoglycerate seems to be increased in proliferating cells (Snell 1984). Serine hydroxymethyltransferase, catalyzing the reversible reaction of serine to glycine, is highly activated in cultures of rat glioma cells (Kohl et al. 1980). Glycine, one product of this reaction, is measurable by proton spectroscopy. It has been shown that glycine is increased in malignant gliomas (Jain et al. 2012; Lehnhardt et al. 2005; Hattingen et al. 2009; Kinoshita et al. 1994; Maudsley et al. 2014). The other product, 5,10-methylene-tetrahydroxyfolate, is utilized for purine and nucleotide synthesis. Therefore, glycine might be considered as a surrogate marker of enhanced glycolysis and nucleotide synthesis.

However, the glycine signal is overlapping with the signal from myoinositol (MI) at 3.56 ppm, requiring special measures for discriminating MI from glycine as described in Chap. Future Methods in Tumor Imaging). Increased MI concentrations or ratios of MI to creatine were detected in tumors, but also in multiple sclerosis, Alzheimer’s disease, and in other metabolic and inflammatory white matter diseases as well as in tuberous sclerosis. Common to all of these pathologies is augmented astrocytic proliferation and demyelinization. Therefore, the role of this metabolite in maintaining cell volume in reactive astrocytes is discussed (detailed discussion and references in Hattingen et al. 2008).

An increased choline signal intensity is frequently observed in 1H MRS data from tumor tissue (Figs. 3a, 4b, and 6) and has been attributed to rapidly proliferating cells (Herminghaus et al. 2002; Guillevin et al. 2008). In conjunction with the decrease of the NAA (N-acetyl-aspartate and N-acetyl-aspartylglutamate) signal intensity due to neuronal loss, the tCho/NAA ratio is considered as the most prominent marker for tumor tissue in MRS (Figs. 1, 3a, 4b, and 6). Modulations of phospholipid turnover, which is in part described by the Kennedy pathway (Kennedy 1957), play a pivotal role in the tumor metabolism (Podo 1999). In brief, this pathway describes synthesis of phosphatidylcholine via choline and phosphocholine (PCho) and its breakdown via glycerophosphocholine (GPC). However, 1H MR spectroscopy detects only total choline (tCho) as the sum of free choline, PCho, and GPC. Consequently, 1H MRS cannot differentiate between PCho and GPC changes, whereas 31P spectroscopy can (Fig. 5). There is increasing evidence that the metabolites PCho and GPC play an important role in tumorigenesis with high PCho/GPC ratios indicating malignant phenotype of a brain tumor (Hattingen et al. 2013).

Fig. 6
figure 6

Diagnostic information from combined long and short TE spectra in high-grade gliomas. Short (30 ms, middle panel and long (144 ms, right panel) TE spectra from glioma grade IV. The upper traces represent tumor tissue from the red-marked voxel in the MRI, while lower traces show contralateral normal-appearing tissue (yellow-marked voxel in the MRI). Note that only the short TE spectrum of the tumor shows a prominent lipid signal at 1.3 ppm. The signal at 3.6 ppm in normal-appearing tissue almost disappears at the long TE, while this signal is clearly visible in tumor tissue for both TE. This indicates that the signal in tumor rather originates from glycine than from myoinositol

Full size image

In vitro studies showed that PCho is the dominant membrane lipid metabolite in proliferating tumor cells and tumor tissues (Gillies et al. 1994). PCho is formed by phosphorylation of choline by the cholinkinase α which is over-expressed in many malignant tumors including glioma cell lines (Glunde and Bhujwalla 2007). Several oncogenes increase choline kinase activity and hypoxia-inducible factor 1 alpha signaling upregulates choline kinase expression (Glunde et al. 2008). Apart from its role as a phospholipid membrane precursor, PCho may also act as a second messenger in cell growth signaling (Gillies et al. 1994; Cuadrado et al. 1993; Aiken and Gillies 1996). Aiken and Gillie 1996 found increased PCho content of rat glioma cells, which decreased during the conversion from the exponential growth to stationary growth phase. Ex vivo MR spectroscopic studies of human brain tumors could further show that the PCho concentration is increased in high-grade gliomas compared to low-grade tumors (McKnight et al. 2011; Vettukattil et al. 2013), but the same studies yielded inconclusive results regarding the GPC concentrations in these glioma specimens. It remains unclear whether low-grade gliomas have higher GPC concentrations compared to high-grade tumors. The amount of GPC might be predominantly influenced by molecular genetic markers and not by the tumor grade. It has been shown that glioma tissue specimens with oncogenic IDH1 mutations have significant higher GPC concentration levels compared to tumors without this mutation (Constantin et al. 2012). Studies in endometrial and ovarian cancers showed that higher activity of the GPC-cleaving enzyme glycerophosphodiesterase increases migration capacity of tumor cells (Papanagiotou et al. 2007). High activity of this enzyme lowers the GPC concentration releasing free choline which can be converted into PCho.

Most of the in vitro results were obtained from tumor cell cultures or tumor xenografts in animal models, representing cells growing as focal mass similar to the majority of body tumors. In contrast, diffuse human gliomas frequently infiltrate large areas of normal brain tissue without significant functional and structural impairment of the host tissue until tumor necrosis and angioneogenesis occur. The prototype of this growth pattern is the gliomatosis cerebri which largely infiltrates different lobes of the brain by sparsely distributed glioma cells in mostly oligo-symptomatic patients. Further, the mitotic rate of human gliomas is quite low compared to most experimental tumor models. Thus, concentrations of metabolites indicating growing tumor cells might be below the detectable limit due to large contributions from regular brain tissue. However, cells from infiltrated normal brain tissue may react to the presence of tumor cells by changing their metabolism too. Some metabolites detected in gliomas by proton spectroscopy may represent this activated non-neoplastic brain tissue. Especially myoinositol and creatine, which are increased in gliomatosis cerebri and some low-grade tumors, seem to be rather markers of reactive gliosis than indicators of typical tumor metabolism (Hattingen et al. 2008).

3 Tumor Grading and Heterogeneity

MR spectroscopy is only one part in the diagnostic work-up of a space-occupying lesion. There is neither a specific tumor metabolite nor a specific spectroscopic pattern which allows unambiguous diagnosis of a glioma. Further, larger glial tumors are commonly heterogeneous with regard to their malignancy and invasiveness, yielding regional-dependent spectral pattern which can be determined with MRSI. Consequently, this is the method of choice to depict tumor heterogeneity which is manifested in heterogeneous distribution of metabolite concentrations (Fig. 1). Further, the location of the MRSI slice can be adjusted to sample the contralateral side providing individual reference metabolite concentrations from normal-appearing brain tissue in the same measurement. In tumors with vast necrotic areas, a vital debris dilutes all metabolites and sometimes only gives rise to large lipid signals which may even spoil the spectral quality (Fig. 2c).

Fortunately, there is a spectroscopic pattern which is very characteristic for gliomas. As already mentioned, gliomas have high tCho signals reflecting higher membrane turnover and cellular density, whereas NAA as marker of viable neuronal tissue is considerably lowered. Thus, drawing a line connecting the tCho peak with the NAA peak normally yields a positive slope, while for non-necrotic high-grade gliomas, the slope is negative (Fig. 4b). Diagnosis of a cerebral tumor is unlikely if the NAA concentration is normal and partial volume effects are excluded (Papanagiotou et al. 2007; Hattingen et al. 2010). Although regular NAA concentrations rule out tumor diagnosis, a decreased NAA signal is not specific for a tumor: NAA is synthesized in neuronal mitochondria and any brain disease severely affecting neuronal tissue can decrease NAA concentration levels. This is especially true for encephalitis or cerebritis, tumefactive demyelinating lesions, and infarction. Similarly, regular tCho signal intensity in all voxels of a non-necrotic space-occupying lesion will exclude high-grade gliomas with high accuracy. But a normal tCho does not exclude any glioma, since glioneuronal tumors, WHO grade II astrocytomas, and gliomatoses often show normal or only slightly increased choline concentrations (Fig. 4a). On the other hand, high tCho signals may be found in pilocytic astrocytomas (Porto et al. 2010), acute brain diseases with high cell membrane turnover like encephalitis, acute demyelinating diseases (Blasel et al. 2011a), active dysmyelination, tuberculosis, and acute radiation injury. Further, lipid signals and lactate are frequently described as tumor metabolites. Lipid signals may occur in tumors without obvious necrosis on conventional MRI, indicating microscopic or even intracellular lipids in high-grade gliomas. However, each of the above mentioned aggressive brain diseases may also yield lipid and lactate signals from necroses and hypoxia. High concentrations of myoinositol and creatine are reported in gliomatosis cerebri and lower-grade astrocytomas, but also in other brain diseases with augmented astrocytic proliferation and demyelinization (detailed discussion and references in Hattingen et al. 2008).

Several studies investigated the accuracy of proton spectroscopy to differentiate between tumors and non-neoplastic lesions and to differentiate low-grade from high-grade gliomas. The differentiation between high-grade and low-grade tumors and differentiation between astrocytoma and oligodendroglial tumors are both decisive for therapeutic decisions. High-grade brain tumors are usually treated more aggressively than low-grade tumors, and higher-grade oligodendroglial tumors are more sensitive to chemotherapy than other tumor entities. A detailed overview and description of these studies is provided by Horská and Barker (2010). The main drawback of the presented studies is the limited comparability due to the differing methodological approaches: SVS versus MRSI, different echo times, different post processing, and various metabolite ratios. Using ratios between different metabolites has the advantage of higher sensitivity if it is obvious that both metabolite concentrations change in the opposite direction. This is the case for the Cho/NAA resp NAA/Cho ratio in neoplastic lesions (Fig. 3a) (Stadlbauer et al. 2007; Vuori et al. 2004; Nelson 2001). However, for some metabolites like creatine and myoinositol, increase and decrease in concentrations were observed. The evaluation of creatine and myoinositol in brain tumors has important diagnostic and also prognostic value. Normally, both metabolite concentrations are decreased in brain tumors. However, elevated creatine and myoinositol levels have been found especially in low-grade gliomas. Higher creatine concentrations compared to normal brain tissue were correlated with shorter progression-free survival (Hattingen et al. 2010). Higher myoinositol levels in brain tumors may support the diagnosis of a low-grade astrocytoma (Castillo et al. 2000), whereas higher glycine concentrations were found in high-grade gliomas as demonstrated in Fig. 6 (Hattingen et al. 2009; Davies et al. 2010).

Alternatively, heterogeneity of a tumor can be evaluated with MRSI analyzing a maximum metabolite level of the tumor related to the same metabolite from the contralateral healthy tissue (Di Costanzo et al. 2008). This approach yields a normalized value which takes interindividual and regional metabolite variations into account. The maximum normalized tCho (hot spot) is also a qualified value for grading non-necrotic gliomas, and the respective voxel might be the target of stereotactic biopsy (Hermann et al. 2008; Senft et al. 2009). The selection of voxel with potentially most malignant tumor tissue is important for tumors in eloquent brain regions which have to be left partially in place.

The peri-enhancing tumor regions should also be sampled and analyzed with MRSI. In contrast to metastases, gliomas infiltrate brain areas beyond the enhancing area, showing elevated tCho concentrations (Fig. 1) and increased Cho/NAA ratios (Fig. 3a) in surrounding tissue (Stadlbauer et al. 2007; Di Costanzo et al. 2008). An investigation of the peri-enhancing border zone has also therapeutic relevance. Considering that all areas of viable tumor have to be targeted with high radiation dose, the “invisible” marginal zone might be undertreated. Recurrent tumors mostly occur in these marginal zones (Blasel et al. 2011b). Thus, integration of MRSI and/or MR perfusion in the treatment planning of high-grade gliomas would target more tumor tissue and might prolong progression-free survival of the patients. This has already been shown for Gamma Knife surgery (Chan et al. 2004).

Although phosphorus spectroscopy seems to be closer to the tumor biology, investigation of tumor heterogeneity or peri-enhancing tumor area is not possible due to its limited spatial resolution. An impression of the rather coarse grid size for 31P MRS can be obtained by comparing the grids in Figs. 4 and 5.

3.1 Some Aspects of Differential Diagnosis

Bearing in mind the above described limitations, MR spectroscopy should only be used in conjunction with MR imaging and age and clinical symptoms of the patient to avoid misdiagnosis. The best diagnostic accuracy can be achieved by combining advanced imaging techniques (Tzika et al. 2003; Chang et al. 2009). Diffusion-weighted imaging is the best method to diagnose an abscess; MR perfusion of the tumor and the peri-enhancing region is highly accurate in grading gliomas and in differentiating infiltrated from focal, non-infiltrating brain tumors (Di Costanzo et al. 2008). Hereby, it is worth to mention that primary CNS lymphomas are also infiltrating brain tumors showing increased blood volume outside the enhancing area (Blasel et al. 2013). Inside the enhancing area of CNS lymphoma, the spectroscopic pattern is “an intermediate” between high-grade gliomas and metastases, showing intermediate tCho increase and prominent lipid peaks at short TE (Harting et al. 2003). The lipid increase might be invisible in long TE MR spectra.

Metastases from different primary tumors show diverse spectroscopic pattern according to their biological heterogeneity. The Cho signal intensity is elevated in solid and proliferating metastases, but most metastases show only moderate Cho increase (Fig. 4b). Huge lipid signals are found in necrotic glioblastomas, but lipids are also the dominant peaks in most of the metastases (Fig. 7) (Poptani et al. 1995). Further, the Cho/NAA ratios of metastases from peritumoral areas differ from the ratios in infiltrating gliomas, indicating the lack of tumor infiltration in the former (Server et al. 2010).

Fig. 7
figure 7figure 7

Monitoring lipids. Typical spectra from the two cases shown as parameter maps in Fig. 3. Lower traces show the normal-appearing tissue (yellow mark in the MRI at left panel) while upper traces show tumor tissue (red mark in the MRI). Note that lipid signals visible at short TE (spectra in middle panel, upper trace) for the glioblastoma (a) are not visible in the long TE spectra (right panel), while for the metastasis (b) spectra for both TE show lipids

Full size image

There are some metabolites which are indicative, but not absolutely specific for special tumor entities (Table 1). Taurine is an organic acid with many fundamental biological roles such as osmoregulation, antioxidation, membrane stabilization, and modulation of calcium signaling. High taurine signal intensities have been found in primitive neuroectodermal tumors (PNET) including medulloblastomas (Panigrahy et al. 2006; Kovanlikaya et al. 2005). Alanine, an amino acid, is found in meningiomas (Poptani et al. 1995; Kugel et al. 1992), but also in abscesses. The spectra of the later typically also show an increase of other amino acids. Multiplets of amino acids (0.9 ppm), lactate (at 1.3 ppm), and alanine (at 1.5 ppm) can be differentiated from lipids by their inversion with a long TE (135–144 ms) (see also Fig. 2 for detection of lactate). Amino acid increase in bacterial abscesses results from enhanced glycolysis yielding high levels of pyruvate, which is the substrate for the amino acid synthesis of alanine and others.

3.1.1 Using Sophisticated Analysis Schemes and/or Pattern Recognition Techniques

Apart from the above-described method of parameterizing MRS data in terms of metabolite concentrations, a different attempt has been made in using pattern recognition techniques for the entire spectrum, determining spectral profiles for each tumor type (Opstad et al. 2007; Tate et al. 1998, 2006).

4 Prognostic Markers

Prognostic markers are applicable to tumors without treatment, whereas in treated tumors only the predictive value of a metabolite can be evaluated. Only few studies with limited patient numbers investigated predictive or prognostic value of tumor metabolites. Multimodal approaches combining different values from various methods may lack of practicality and comparability between institutions. The impact of most spectroscopic studies in this area is limited by partial or even total lack of histopathological confirmation. Histopathologically proven studies showed that monitoring a tumor with MR spectroscopy may increase sensitivity and specificity to detect tumor progress or malignant transformation (Rock et al. 2002). Tedeschi et al. reported a continuous increase in the tCho signal to the time point of malignant transformation in low-grade tumors (Tedeschi et al. 1997), and Graves at al. found a tCho increase in recurrent malignant gliomas after Gamma Knife radiosurgery (Graves et al. 2001). But one should keep in mind that transient tCho increase might also occur in the radiated brain tissue.

As already mentioned, high normalized creatine concentrations in untreated WHO grade II and III gliomas are correlated with shorter progression-free survival. The role of creatine in glial tumors is unknown. Most spectroscopic studies used metabolite ratios related to creatine, which lacks information on the real creatine concentrations. No creatine increase was found in glioma cells ex vivo, suggesting that the increase rather originates from (reactive) glial cells of the infiltrated brain. Further, as the creatine signal in 1H spectra represents the sum from unphosphorylated and phosphorylated creatine, the information on tumor energy metabolism obtained from intensity changes of this signal is limited and relies on additional assumptions regarding its composition and compartmentalization (Hattingen et al. 2010).

5 Treatment Monitoring

The main drawback of proton spectroscopy in treated high-grade gliomas is the small fraction of viable and solid tumor tissue in a brain area with sufficient field homogeneity to provide artifact-free spectra. Almost all patients are treated with radiation, and most patients receive at least one chemotherapeutic regime. Therefore, most lesions are heterogeneous consisting of both progressive tumor and a considerable amount of pre-injured tissue (Rock et al. 2002). In our experience, a large amount of spectroscopic data do not match the criteria for spectral quality (Kreis 2004) to allow a reliable analysis of the metabolite concentrations. The same seems true for distinguishing pseudoprogression and true tumor progression. The so-called pseudoprogression is regarded as intense reaction to combined radiochemotherapy, which decreases without additional treatments thereafter. Until now, MR spectroscopy was not very successful in differentiating pseudoprogression from real progression (Hygino da Cruz et al. 2011).

Therapy-induced brain injuries occur in about 20–30 % of patients treated with temozolomide radiochemotherapy. These lesions enhance early after radiation which may imitate tumor progression (Brandes et al. 2008). For adequate therapy decisions, additional methods are required to differentiate these reactions from real tumor growth.

Phosphorus spectroscopy might be the more appropriate method for treatment monitoring, since it is less prone to artifacts and, although of inferior spatial resolution, could be more specific by differentiating between the phosphomonoesters and phosphodiesters. First data on a cohort of patients with recurrent glioblastomas, all treated with bevacizumab in the second line, yielded that PCho/GPC seems to be appropriate to predict survival time and also to detect tumor progress (Hattingen et al. 2013).