Mapping of Glycine Distributions in Gliomas

Abstract

Background and Purpose

Increased glycine concentration in the brain is associated with altered metabolism in cancer and can be detected using in vivo MR spectroscopy. This has been proposed as a marker for grade IV gliomas; however, little is known about the potential significance and frequency of in vivo glycine observation. The purpose of this study was to examine the rate of occurrence and spatial distribution of glycine observation with respect to other MRI parameters.

Materials and Methods

Data from volumetric whole-brain MR spectroscopic imaging of 59 subjects with glioma were analyzed with glycine included in the spectral model. The associations of the signal amplitude and spatial distributions of glycine with findings from contrast-enhanced T1, perfusion, and diffusion MRI were then examined.

Results

Glycine was detected in 24% of all studies, although with a wide range of signal amplitude and extent of the spatial distributions. While more commonly seen in grade IV tumors (42% of studies), relatively large concentrations were also detected in grade II and III gliomas. Co-analysis with other metabolites indicated a strong association with choline and that glycine was frequently seen to be overlapping with, and adjacent to, areas of high lactate concentration. Increased glycine was always associated with contrast enhancement and areas of increased cerebral blood flow, but without any clear association with other image parameters.

Conclusions

Detection of increased glycine in gliomas appears to identify a sub-group of tumors and to identify areas of increased proliferation.

Introduction

Glycine (Gly) is an inhibitory neurotransmitter and key amino acid that acts as an intermediary for synthesis of nucleotides and glutathione, and is present in normal human brain at up to 1 mM concentration (1). Increased concentrations of Gly have been detected in brain tumors using in vivo and ex vivo MR spectroscopy (2–8) and it has been proposed as a marker for grade IV glioma (6). However, in vivo studies have been limited, and the relative frequency of observation, distribution, association with other imaging measures, and potential value as a marker of glioma grade have not been demonstrated.

The use of Gly as a tumor marker is limited by its relatively low concentration and difficulty for in vivo detection of the singlet resonance at 3.55 ppm due to overlap with a myo-Inositol multiplet that appears as a prominent peak in the same spectral position. Several studies have demonstrated that improved discrimination can be obtained by using long TE acquisitions that diminish the mI signal contribution through the J-coupled evolution of the multiplet structure (1–3, 5), and by combining results from short- and long TE values the two signal contributions can be discriminated. The ratio of the combined mI+Gly peak at two TE values has been proposed as a marker of malignancy for tumor grading (3, 5). These previous studies have presented summary measures that indicate the potential for discrimination between patient groups; although with a considerable range of the measured mI+Gly signals within each subject group. However, the frequency of increased Gly detection has been reported in only one study by Choi et al., where the Gly concentration was increased in 8 out of 12 subjects with glioblastoma multiforme (GBM), again with considerable variability between subjects.

While previous reports have indicated potential for Gly measurement as a molecular marker to aid in tumor classification and grading, the significance of increased Gly, the relative frequency of in vivo MRS observation, the reliability of the measurement, and the potential role as a diagnostic marker remain poorly understood. In this study an analysis of volumetric ¹H MRSI of the brain in subjects with glioma has been carried out to examine the relative frequency of increased Gly in this subject group and to examine the distributions of increased Gly in relation to other imaging measures, including diffusion and perfusion MRI.

Methods

This retrospective analysis was applied to data acquired for a group of 59 subjects aged 14 to 69 with histologically-proven glioma with World Health Organization classifications of grade II (20 subjects), III (15 subjects), or IV (24 subjects), prior to treatment. Subjects underwent a MRI study at 3 Tesla (GE Signa HDxt) that included a whole-brain MRSI, DTI, DCE perfusion measurement, T2-weighted MRI, and pre- and post-contrast T1-weighted MRI, for a total study time of 50 minutes. Details of the subject selection and imaging methods have been previously reported (9) and therefore will be described only briefly.

The MRSI acquisition used a volumetric Echo-Planar acquisition covering the cerebrum with an effective voxel volume of approximately 1 mL. The sequence used spin-echo excitation with TE=70 ms and inversion-nulling with TI=198 ms to suppress subcutaneous lipid and macromolecular signal contributions. At this TE value the signal from mI at 3.55 ppm is significantly reduced due to J-coupling evolution, thereby improving spectral analysis of Gly, as has been demonstrated at different field strengths in previous studies (7, 10). This is illustrated in Fig. 1a using spectral simulation (7) of equal concentrations of Gly and mI at 3 T for a line width of 7 Hz. The signal from phosphocholine is also shown as a reference. In Fig. 1b is shown the signal integral of the real part of the mI signal for an 11 Hz region centered at the position of the Gly resonance. It can be seen that at this TE value the mI signal contribution at 3.55 ppm is considerably reduced, thereby improving detection of Gly using a resonance defined at this position. The mI spectral pattern is also quite distinct and can be visually discriminated from that of Gly. MRSI processing was carried out using the MIDAS package (11, 12), which included automated spectral analysis for NAA, tCre (creatine and phosphocreatine), tCho (glycerophosphocholine, phosphocholine, and free choline), Gly, and Lac. With this signal model there remains a possible signal contribution from mI to the fitting of the Gly singlet resonance; however, for all analyses a visual inspection of the spectra was used to test for the presence of mI, which could be distinguished by the positive and negative peak combination at 3.66 and 3.5 ppm. Similarly, although the lipid inversion-nulling preparation pulse also reduces any mobile lipid signal from tumors some contribution to the Lac signal fitting cannot be excluded and this result is therefore referred to as lipid+lactate (LL). The metabolite maps were signal normalized using tissue water, which was obtained as a second SI acquisition interleaved with the metabolite SI acquisition.

Figure 1.

(a) Simulation of the spectrum from mI, Gly, and phosphocholine for spin-echo observation at TE=70 ms and 3 Tesla. In (b) is shown the signal integral of mI over an 11 Hz region centered at the position of the Gly resonance at 3.55 ppm.

To improve detection of small Gly signal contributions the MRSI spectral analysis was performed using a two-stage approach similar to that proposed by Zhang et al. (13). The MRSI data was first analyzed following application of spatial smoothing using convolution by a spheroidal kernel of diameter 5×5×3 voxels, equivalent to 4.2 mL. The resultant signal amplitudes from this fit were then used as initial values for analysis of the original MRSI data following correction for phase, B0, and baseline variations. Metabolite images from both analysis results were available for the subsequent analyses.

DTI data were acquired using a single-shot echo-planar sequence, with TR=10 s, TE=100 ms, 46 slices of 3 mm thickness, FOV= 240 mm, image matrix = 128*128, diffusion weighting b-factor of 1000 s/mm² applied in 12 directions in addition to the b=0 s/mm² measurement, for an acquisition time of 2 min. 34 sec. Data was processed using in-house developed software to compute the mean diffusivity and fractional anisotropy for each voxel.

DCE-MRI was performed using a 3-dimensional SPGR sequence (TR/TE/flip angle = 5.0 msec/2.1 msec/10°; slice thickness 6 mm), with 32 time points for 12 slices spanning the Gd-DTPA-BMA (Omniscan) administration (5 mL/s and 0.2 mmol/kg body weight). Total acquisition time was 10 min. Quantitative analysis of the concentration-time curve was performed to calculate the CBF and a corrected CBV map that removed the contrast agent leakage effect due to the disrupted BBB (14).

The tumor and surrounding edema were manually outlined using the coregistered T1- and T2-weighted MRIs to create a mask that was identified as the gross tumor volume. This was then resampled to correspond to the MRSI resolution and spatial response function. This, and all metabolite, DTI and DCE maps, the T2-weighted MRI, and the post-contrast T1 MRI were then spatially transformed into a standard reference space at 2 mm isotropic resolution.

Two types of data analysis were performed to examine associations between the relative Gly concentrations and values from the other image measures. Firstly, data points for all image types were obtained from voxels within the region defined by the gross tumor volume mask and Pearson correlation coefficients determined between Gly and all other image measures. In consideration of the interpolation applied for spatial normalization of the metabolite images only every third voxel (6 mm spacing) was taken in order to limit the number of data points. Voxels were also excluded if the spectral fitting reported a linewidth of greater than 12 Hz. The second type of analysis was to qualitatively view the relative spatial distributions of increased Gly values relative to features in the other metabolite images and MRIs. For those studies where Gly concentrations were sufficiently visible above background levels a contour was drawn to indicate the Gly distribution that was then superimposed on other images of interest, with care taken to maintain registration of the tumor volume relative to the edge of the brain. To estimate Gly concentrations the Gly integral from the spectral fitting was scaled by the corresponding value for tCre in contralateral white matter and assuming a tCre concentration of 8 mM (4). Where possible, mean values were obtained over 6 to 10 contiguous voxels in the highest amplitude region of the Gly image and from a similar-sized region in contralateral white matter.

Results

Out of 59 datasets, 14 (24%) MRSI studies were found to exhibit a visible Gly signal, with 1 of 20 (5%) grade II, 3 of 15 (20%) grade III, and 10 of 24 (42%) grade IV gliomas. However, the concentrations and extent of the Gly signal varied considerably, from barely detectable in a small location (5 subjects), to a prominent signal corresponding to an estimated maximum concentration of 9.6 mM in a grade III astrocytoma. Of those studies that exhibited Gly, the average of the maximum estimated Gly concentrations were 6.4 mM (range 3.4 to 9.6 mM) for the low grade tumors and 3.6 mM (range 1 to 6 mM) for the high grade; however, concentrations within each tumor varied considerably. The increased Gly was directly visible as a hyperintense region in the metabolite maps of 7 studies (11.9%), whereas identification in the remaining studies required visual inspection of individual spectra, for example in cases where only isolated voxels showed an elevation or a small increase in concentration required confirmation by signal averaging over a region. In Fig. 2 are shown example spectra and spectral fit results for a grade III astrocytoma, with an Gly concentration (estimated relative to contralateral white-matter tCre) of 6 mM, and a grade IV glioma with estimated concentration of 3 mM. These results demonstrate good quality fitting for all metabolites with minimal impact of mI at the 3.55 ppm resonance of Gly within the signal-to-noise limitations of the MRSI acquisition. The Cramer-Rao bounds (CRB) for the Gly analysis were 6% and 19% respectively.

Figure 2.

Example single-voxel spectra and spectral fitting results for: a) a astrocytoma grade III, and b) a GBM. Estimated concentrations of Gly are 6.0 and 3.0 mM, with CRB values of 6% and 19% respectively.

In Fig. 3 are shown representative metabolite images at multiple slices for a single subject with a GBM. This illustrates the quality of the volumetric metabolite maps, which also include sampling close to the cortical surface. This result shows widespread loss of NAA over a region comparable to the extent of T2 signal enhancement and a ring of increased Cho corresponding to tumor tissue surrounding a necrotic core. This study showed a relatively small increase of Gly, estimated at a maximum of 3.7 mM concentration, and Fig. 2e shows the Gly map obtained from fitting of the spatially-smoothed MRSI result. The CRB for the Gly analysis was ~8% in the area of maximum signal. The Gly map shows increased signal over just one section of the tumor, in a region that also shows increased Cho and heterogeneous post-contrast T1 signal enhancement. The LL signal is seen on lower slices, distributed within both the necrotic core and the area of increased Cho, and is displaced laterally and vertically from the area of increased Gly.

Figure 3.

Example volumetric MRSI data for a subject with a GBM, showing a) T2-weighted MRI; b) post-contrast T1-weighted MRI; c) NAA; d) Cho; e) Gly, and f) LL.

In Fig. 4 are shown multiparametric images at 2 slices from a grade II (Fig. 4a) and a grade IV (Fig. 4b) glioma, together with example spectra showing the presence of the Gly peak at 3.55 ppm and a representative spectrum from normal appearing white matter. A contour based on the Gly distribution at each slice has been superimposed on each of the parametric maps. In the grade II example the Gly region corresponds closely to the solid contrast-enhanced tumor region and to increased CBF. In this case it can be seen that the area of increased LL overlaps with that of increased Gly, although the strongest LL signal is outside of the Gly region. For the grade IV tumor the increased Gly corresponds very closely to the regions showing decreased NAA and increased Cho and CBF. In both cases there are no image features in common with the MD image, and similarly the FA image (not shown). Both tumor spectra show a clearly identifiable Gly signal at 3.55 ppm, plus that of the grade II astrocytoma shown in Fig. 4c shows a strong signal from the out-of-phase Lac doublet centered at 1.3 ppm together with a contribution from the alanine doublet, centered at 1.47 ppm.

Figure 4.

Example visualizations of glycine distributions in relation to other MR image parameters, for: a) a grade II astrocytoma and b) a grade IV glioblastoma. The MR parameter is indicated at the top of each column of images. Example spectra from the grade II tumor are shown in (c), and from the GBM in (d). The spectrum shown in (e) is from a normal-appearing white matter region of the GBM subject. T1-PC is the post-contrast T1-weighted MRI and MD is the mean diffusivity image.

All studies with visible Gly showed T1 contrast enhancement and the presence of LL. For the high grade tumors Gly was commonly seen in only a small portion of the tumor, which always corresponded to heterogeneously enhancing regions, with areas of enhancement mixed with no enhancement (e.g. Fig. 3), and to increased CBF and CBV. The result for the grade II tumor shown in Fig 3a was the only case where increased Gly was seen throughout the solid enhancing region. Gly was not seen in areas of necrosis, in contrast to LL which could in general be seen throughout the enhancing and necrotic regions.

In six of the studies where Gly was observed the signal was present in only a small portion of the gross tumor volume, and therefore these studies were not included in the regression analyses. The Pearson correlation coefficients and confidence intervals for the remaining eight subjects are presented in Table 1 for selected parameters that showed the most significant findings. Considerable variability across subjects is demonstrated, although all studies showed a positive correlation of Gly and tCho, and four showed a moderate-to-strong negative association of Gly with NAA. Results for Lac, CBF, and CBV showed mixed results, and there were no strong associations with any of the diffusion parameters.

Table 1.

Correlation coefficients (R) and 95% confidence intervals (CI) for the linear regression of Gly with selected MRS and MR perfusion parameters.

Study ID	Age	Tumor Type	R (95% CI)
			NAA	tCho	Lac	CBF	CBV
07	40	GBM	−0.71 (−0.76, −0.64)	0.07 (−0.04, 0.19)	0.64 (0.57, 0.71)	0.37 (0.27, 0.47)	0.30 (0.19, 0.40)
25	30	GBM	−0.44 (−0.51, −0.36)	0.63 (0.57, 0.69)	0.34 (0.25, 0.42)	0.45 (0.37, 0.52)	0.41 (0.33, 0.49)
39	38	GBM	0.07 (−0.05, 0.19)	0.37 (0.27, 0.47)	−0.25 (−0.36, −0.13)	0.05 (−0.07, 0.17)	0.11 (−0.01, 0.22)
44	20	Astrocytoma	−0.16 (−0.25, −0.07)	0.49 (0.42, 0.55)	−0.05 (−0.14, 0.04)
59	36	Astrocytoma	−0.55 (−0.60, −0.50)	0.15 (0.08, 0.22)	0.18 (0.11, 0.26)
68	41	GBM	0.21 (0.13, 0.30)	0.24 (0.16, 0.32)	−0.15 (−0.23, −0.07)	0.01 (−0.08, 0.09)	−0.02 (−0.11, 0.07)
72	65	GBM	0.12 (−0.07, 0.30)	0.20 (0.02, 0.37)	−0.03 (−0.22, 0.15)	0.04 (−0.15, 0.22)	0.11 (−0.08, 0.29)
78	14	Astrocytoma	−0.33 (−0.45, −0.20)	0.79 (0.73, 0.84)	0.11 (−0.03, 0.24)	0.16 (0.02, 0.30)	0.21 (0.07, 0.34)

Discussion

This study has been the first to apply a volumetric MRSI acquisition to map distributions of Gly in gliomas and to examine potential associations of increased Gly distributions with other MRI parameters. The findings of this study include that increased Gly is seen in approximately a quarter of the studies, with a greater frequency of occurrence in grade IV tumors, although not specific to high grade gliomas. In most of the cases the Gly signal was not distributed throughout the tumor volume and was associated with regions of T1 contrast enhancement and increased CBF. By using a regression analysis it was found that there is a strong association of Gly concentration with that of Cho.

The primary pathway for Gly synthesis in the brain is through conversion of glucose via 3-phosphoglycerate and serine (15–17). Measurements in cultured cells have indicated that the rate of glucose consumption leading to Gly production differs among cell lines, but can almost match that of Lac in melanoma cells (18). This pathway is initiated by phosphoglycerate dehydrogenase, which is encoded by the gene PHGDH and has been associated with increased mortality in breast cancer (16), melanoma (19), and glioma (17). Lie et al. found that PHGDH was increasingly expressed in more aggressive glioma types, and Jain et al. (16) reported that Gly consumption in cancer cell cultures is associated with rates of proliferation and that it is released by slowly proliferating cells, suggesting that Gly may build up in areas where synthesis exceeds demand. These previous findings suggest that while Gly observation is an indicator of this specific genetic marker, it additionally reflects the balance between production and consumption and the influence of the tumor microenvironment. It therefore remains to be shown whether in vivo Gly observation may provide a reliable marker of a genetic mutation affecting PHGDH expression, and future studies examining the association between Gly distributions, histology, and PHGDH expression are recommended to examine this question.

The relative number of studies in which Gly was detected in GBMs (42%) was smaller than that of 66% reported by Choi et al. (4), and the 87% of glioma samples with high PHGDH expression reported by Liu et al. (17). Reasons for this include the lower signal-to-noise ratio of the MRSI acquisition relative to the single-voxel measurement and limitations of the MRS acquisition for sampling tumors located in inferior temporal and frontal brain regions, which were included in this study. However, this study also demonstrates a limitation of using single-voxel acquisitions in that for a third of the studies the increased Gly signal was found in only a small portion of the tumor, which could be missed when using a single localized measurement.

The finding of an association between Gly and Cho is consistent with the report of Jain et al. (16) that found a significant correlation between phosphocholine and glycine and that Gly consumption is specific to rapidly proliferating cells. The indirect association of Gly with Lac, with the Gly signal being seen on the edges of areas of increased Lac, as well as the association with areas on increased CBF and CBV, would suggest that Gly production is greatest in regions with incomplete hypoxia.

Limitations of this study include that mI may still contribute to the spectral fitting of the Gly signal and strict quality criteria based on CRB values were not used for the identification of increased Gly in the metabolite maps. To minimize the possibility of mI contributions a visual examination of individual spectra was always used to confirm the appearance of an in-phase singlet resonance. Previous studies have provided detailed analyses for the “optimium TE” approach for different field strengths (7, 10). Improved spectral quantitation of Gly is possible using longer TEs and multi-echo detection sequences (3–5), although at the expense of reduced signal intensity, which would impact the quality of MRSI measurements.

In summary, this study has shown that in vivo mapping of increased Gly in gliomas can be achieved using MRSI, but that it can be detected in only a fraction of subjects and the concentrations and distributions can be highly variable. While previous studies indicate an association of increased Gly expression with a tumor subtype having a specific genetic mutation, the ability to detect the signal in vivo is likely modulated by additional factors from the tumor microenvironment.

ABBREVIATIONS

tCho

total choline

tCre

total creatine

DCE

Dynamic Contrast-Enhanced

Gly

glycine

Lac

lactate

LL

lipid+lactate

MD

mean diffusivity

mI

myo-Inositol

MRSI

Magnetic Resonance Spectroscopic Imaging

NAA

N-Acetylaspartate

ppm

parts per million

SPGR

spoiled gradient recalled echo