³¹P T₂s of phosphomonoesters, phosphodiesters, and inorganic phosphate in the human brain at 7T

INTRODUCTION

In vivo phosphorous-31 (³¹P) MR spectroscopy provides an in situ window on cell energy and phospholipid membrane metabolism. Because these processes are crucial in the development and sustenance of a healthy brain function, ³¹P MR spectroscopy has been applied in the investigation of the healthy brain 1-3 and brain diseases such as schizophrenia 4, 5, bipolar disorder 6-8, Alzheimer's disease 9, 10, attention deficit hyperactivity disorder 11, Parkinson's disease 12, epilepsy 13, 14, multiple sclerosis 15, 16, and brain tumors 17-19.

Although ³¹P MRS on clinical 1.5T MR systems is hampered by low signal-to-noise ratio (SNR) and low spatial, temporal, and spectral resolution, the development of human high field (3T) and ultrahigh field (7T) systems alleviates these restraints. However, the increased spectral resolution at 7T impairs the chemical shift range over which homogeneous radiofrequency (RF) excitation and refocusing can be performed. Additionally, ultrahigh field MR leads to more RF deposition as compared to MR at lower field strengths.

The first in vivo ³¹P study of the human brain at 7T by Lei et al. 1 was published in 2003. In that study, the T₁ and T₂ of phosphocreatine (PCR) and adenosine triphosphate were determined. Accurate knowledge of MR relaxation parameters is essential in optimizing spectroscopy pulse sequences for repetition time and echo time (TE), and is required for absolute quantification of metabolites. In addition, relaxation properties may provide information about the molecular dynamics of metabolites and tissue susceptibilities.

Here we report on the T₂s of the ³¹P metabolites in the chemical shift range of 2.9 to 6.9 parts per million (ppm), that is, the phosphodiesters (glycerophosphocholine (GPC) and glycerophosphoethanolamine (GPE)), inorganic phosphate (Pi), and the phosphomonoesters ((phosphocholine (PC) and phosphoethanolamine (PE)), using multi-echo acquisitions with incorporation of imperfections in flip angles. To the best of our knowledge, this is the first report on the T₂s of these metabolites in the human brain at 7T.

In previous work, we determined the T₂s of these metabolites in healthy breast tissue 20 and breast cancer tissue 21 at 7T. Of particular interest in these studies was a much shorter T₂ for Pi in breast cancer tissue than in healthy breast tissue, which we attribute to upregulated glycolysis in breast cancer tissue, causing reversible exchange between Pi and adenosine triphosphate (ATP). Because the glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA expression of the human cortex is among the tissues with the highest expression 22, and the GAPDH enzyme mediates this Pi–ATP exchange 22, we hypothesized to observe a short T₂ of Pi in the brain as well.

METHODS

Nine healthy volunteers who gave written informed consent were scanned on a 7T whole-body MR scanner (Philips, Best, The Netherlands) with a dual-tuned ³¹P-¹H head coil (MR Coils BV, Zaltbommel, the Netherlands). The scan protocol consisted of basic imaging for positioning, B₀ map for image-based shimming with a shim tool (MR Code BV) based on FASTERMAP 23, and a ³¹P flip-angle series (block pulses with step 15 degrees) in the nominal range 315 to 420 degrees to calibrate the flip angle in obtaining an actual flip of 360 degrees. After B₀ shimming and ³¹P pulse calibration, a multi-echo 3D chemical shift imaging scan (Fig. 1) with spherical k-space sampling was performed (repetition time = 6 s; field of view = 280 × 280 × 280 mm³; matrix = 8 × 8 × 8; bandwidth = 7,000 Hz; 256 data points on a full echo; ΔTE = 45 ms; amplitude of (excitation) RF field (B₁) = 40 μT), acquiring one free induction decay (FID) and seven full echoes. Proton decoupling was not used because the in vivo spectral line width at 7T is larger than the scalar coupling constants.

The performance of the implemented multi-echo sequence, with composite block pulses for refocusing, was simulated using TopSpin 3.5 (Bruker, Billerica, Massachusetts, USA) and compared to a simulated multi-echo sequence with 180-degree block pulses for refocusing. Simulations were done without taking into account transverse or longitudinal relaxation effects. Deviations from perfect refocusing at the chemical shifts of the metabolites of interest were modeled with exponential decay. The simulated signals S_n,m of the n echoes at chemical shift of metabolite m, were fitted to

(1)

where T_2,pp,m is a time constant that is dependent on the RF pulse performance (pp) at the chemical shift of metabolite m. With the T_2,pp,m obtained from simulations, the measured experimental volunteer data can be fitted to

(2)

The metabolite T_2,m and the pp T_2,pp,m can be combined to an apparent metabolite T_2,m,apparent

(3)

Acquired ³¹P data were spatially Hanning-filtered; FID data was first-order phased by circular shifting the first missing data points resulting from the delay between pulse and acquisition. All data were zero-filled in the time domain. After Fourier transform, the spectra were zero-order phase-corrected (based on the absolute full echo signal), and a baseline correction was applied to the FID spectrum to correct for the baseline rollover artefact as a result of the first missing data points. For each volunteer, one voxel was chosen for analysis. Because the nominal voxel size is 35 × 35 × 35 mm³, the effective voxel size after Hanning filtering spans a large part of the brain. A voxel center location was central in the left–right and anteroposterior directions of the brain and a few centimetres below the skull in a feet-head direction. This corresponds to the voxel with the highest SNR for optimal T₂ analysis. Spectral fitting was performed with jMRUI 5.2 24 using the AMARES algorithm 25. Line widths of PE, PC, GPE, GPC were kept identical and free; and the line width of Pi was free. All metabolites were fitted using Gaussian line shapes. Chemical shift of metabolites was fixed within a 0.3 ppm range. After spectral fitting, the calculated amplitudes of the metabolites within FID and echoes were fitted to Equation [2]. The Cramér–Rao lower bounds of the spectral fitting were used as uncertainties in the exponential fitting with Equation [2]. The fitted T₂ values per metabolite and per volunteer were variance-weighted to obtain a best average T₂ value per metabolite. To enable the acquisition of seven echoes, the sampling times for FID and echo acquisitions were kept relatively short, leading to a small truncation artefact in the spectra. This artefact was removed (for viewing purpose only) by replacing the truncated tail of the experimental time domain data by the tail of the fitted time domain data obtained from jMRUI 5.2.

RESULTS

In Figures 2a,2b, the results for the simulations of the multi-echo sequence with 180-degree refocusing block pulses and refocusing composite block pulse are compared, respectively. It is clear that with 180-degree block pulses at B₁ = 40 μT, the first echo already shows a drop in intensity to approximately 90%, within the 4-ppm range of interest of approximately 4.7 ppm. A better refocusing performance is seen with the composite block pulse, although a small drop in intensity can be observed, particularly on the edges of the 4-ppm range of interest.

Applying the multi-echo sequence with composite block refocusing pulses will lead to a small underestimation of the T₂ of metabolites, which can be corrected by taking into account the pp profile of Figure 2b, for which the deviation can be fitted as an exponential decay given by Equation [1]. The results of fitting the corrective pp time constants, T_2pp,m, at the different chemical shifts of the metabolites, are shown in Figure 3.

The numerical values of the pp time constants T_2,pp,m in seconds, which were fitted at the different chemical shifts, are summarized in the second column of Table 1. An example of the chosen voxel localization for analysis, a full FID spectrum, and the multi-echo spectra in the chemical shift range 2 to 8 ppm for one volunteer are shown in Figure 4.

Because the refocusing bandwidth of the composite pulses is limited to approximately 4 ppm, only a small chemical shift range of the ³¹P spectrum spanning the phosphodiesters and phosphomonoesters is shown for the multi-echo spectra. The T₂ fits of the different ³¹P metabolites obtained from fitting the spectra at the different TEs for this volunteer are shown in Figure 5.

The truncation artifact that arises due to the relatively short TE (45 ms) and subsequent short sampling time (18.3 ms for the FID and 36.6 for a full echo) is illustrated in Figure 6. Figure 6a shows the spectrum of an almost fully sampled FID (55 ms sampling); Figure 6b shows the same spectrum but truncated after 18.3 ms (zero-filled to the original size); and Figure 6c shows the partial removal of the artifact (for viewing purpose) by using the tail of the FID as obtained by spectral fitting in jMRUI 5.2.

Averaged multi-echo spectra for the group of nine volunteers are shown in Figure 7.

An overview of all metabolite calculated T₂ values with the standard deviation of the fit for the nine subjects is shown in Figure 8. The last column in the graph shows the variance-weighted average of the metabolite T₂ over the nine volunteers. Numerical values of the variance-weighted averages of the metabolite transverse relaxation constants, T₂, are summarized in the third column of Table 1.

DISCUSSION

The determined time constants for correction of the pulse performance of the refocusing composite block pulse are in the range of 2.3 s to 14.4 s at the chemical shift of the metabolites investigated. With an intrinsic metabolic T₂ of 200 ms, this amounts to a correction that is in the range of 1% to 8%; however, with an intrinsic metabolic T₂ of 100 ms, the correction is only 1% to 4%.

The spectra derived from the FID and echo acquisitions, as depicted in Figure 4, clearly show signals from PE, PC, intracellular Pi (Pi_in), GPE, and GPC; however, there appear to be a few more signals visible. Due to the truncation artefact, it is difficult to distinguish between real signals and artefacts. The phosphomonoester signals (PE + PC) of the FID spectrum seem somewhat broadened. It is possible that two signals from 2,3-diphosphoglycerate (2,3-DPG), which is abundantly present in red blood cells, could contribute to the signal of PC, and to a lesser extent to PE in the FID spectrum. The visibility of these two ³¹P signals from 2,3-DPG in the brain has been previously inferred 26, 27. The chemical shift of the 2,3-DPG signals is highly dependent on the intracellular pH and the oxygenation of the red blood cells. For 7.3 ≤ pH_i ≤ 7.4 and 1% ≤ oxygenation < 98%, the chemical shift (referenced to PCr) of the 2P signal ranges from 5.4 to 6.2 ppm, whereas the chemical shift of the 3-phosphorous signal ranges from 6.2 to 6.9 28-30. This spans the range of the phosphomonoesters PC and PE, with PC approximately 6.25 ppm and PE approximately 6.8 ppm. From a concentration perspective, 2,3-DPG from blood can contribute significantly to the signal of PC, although the impact of 2,3-DPG will probably be limited to the FID signal and possibly the first echo because the blood flow and the spoiler gradients around the refocusing pulses will dephase these signals substantially in the later echoes. Cerebral blood volume in healthy adults is approximately 3.8% of cerebral volume 31, whereas the normal 2,3-DPG concentration in packed red cells is 4.5 to 5.1 mM 32 and in normal haematocrit (males and females combined) is 36% to 50% 32. Combined, this leads to a cerebral 2,3-DPG concentration ranging from 0.061 to 0.099 mM. With two ³¹P signals for one molecule of 2,3-DPG, the contributing signal to the PME signals is in the range of 0.12 to 0.2 mM. From autopsies, it is known that the normal concentration range of PC in the brain is 0.30 to 0.40 mM 33. Consequently, the PC signal as obtained by ³¹P MRS could be 67% too high (upper limit) due to the contributing signals from 2,3-DPG.

The signal labeled Pi_in has a left shoulder of approximately 5.3 ppm that could be extracellular Pi. Extracellular phosphate in the brain has been identified previously by Ren et al. 2, who also measured the T₁ of the intracellular and extracellular phosphate signals in the brain. The origin of the signal at 4.3 ppm is not clear and is not confined to the spectra of the single volunteer in Figure 4. On averaging the spectra from all volunteers, this signal remains visible (Fig. 7). We are inclined to identify it as a truncation artefact, as shown in Figure 6; however, Pi in an acidic environment of pH of approximately 6.7 could possibly also explain the signal. Low pH values are not uncommon in, for instance, lysosomes; and a series of eight Pi signals at different pH have also been reported in the normal isolated dog brain, spanning the chemical shift range of 4.07 to 5.17 ppm 34.

The T₂s of PE, GPE, and GPC that were determined in the brain at 7T are comparable and approximately 200 ms. The apparent T₂ of PC is most likely too low because it is possibly influenced by the 3P and 2P signals of 2,3-DPG from blood, which have a shorter apparent T₂ than PC (due to blood flow and spoiler gradients around the refocusing pulses), as can be seen from the subsequent echo spectra in Figures 4 and 7. Upon omitting the FID and first echo signal for the T₂ fits of PC, a variance-weighted T₂ of 182 ± 7 ms is found. Fitting the T₂ of PC with a two-component model, taking into account a variable fraction of non-PC signal with a variable T₂, leads to a variance-weighted T₂ = 203 ± 4 ms for PC. Here, we first determined an average T₂ and an average fraction for the underlying non-PC signal for the group of nine volunteers by fitting the individual PC data with the following constraints: 140 ms ≤ T₂ (PC) ≤ 350 ms; fraction non-PC signal between 10% and 67%; and 0 ms ≤ T₂ (non-PC) ≤ 75 ms. This leads to a fraction of non-PC signal of 52% and T₂ (non-PC) = 37 ms when averaged over all nine volunteers. These values were subsequently used, and fixed, to fit the individual T₂ data of PC for the nine volunteers. Variance weighting of these individual T₂s of PC then leads to the variance-weighted T₂ value of 203 ± 4 ms. These values are more in line with the T₂s of PE and the phosphodiesters.

Phosphorus spectroscopy, utilizing polarization transfer, indeed shows a much cleaner signal in the phosphomonoester range in the human brain because the underlying non-PC and non-PE signals are not amplified, as compared to a direct detection acquisition 35, 36.

The rather short T₂ of Pi_in of only 86 ± 2 ms is likely caused by exchange with γ-ATP via the reversible reactions involving the GAPDH and phosphoglycerate kinase enzymes of the glycolytic pathway. It is known that GAPDH mRNA expression of the human cortex is high 22. The short T₂ time of Pi_in is corroborated by a short T₁, which was measured by Ren et al. 2, who found a 40% shorter T₁ for intracellular Pi as compared to extracellular Pi in the healthy human brain at 7T. These authors also attributed the shorter T₁ of Pi_in to exchange with γ-ATP. In addition, saturation transfer measurements in rat brain at 11.7T revealed that the extracellular inorganic phosphate Pi_ex is insensitive to saturating γ-ATP, whereas the signal of the intracellular pool Pi_in is reduced, designating a cytosolic origin for exchange between Pi and γ-ATP 37. Note that saturation transfer measurements in skeletal muscle, with the aim of measuring the mitochondrial Pi to γ−ATP flux, always lead to extreme high values as compared to other methods 38. Here also, the cause is sought in the exchange between Pi and γ-ATP through the glycolytic pathway 38. Indeed, the T₂ of Pi_in from skeletal muscle is much shorter than, for instance, the T₂ of the phosphodiesters or PC 39, 40 signals from skeletal muscle. Recently, we measured the T₂ of ³¹P metabolites in healthy breast tissue and breast cancer tissue. For Pi in healthy breast tissue, we found 180 ± 4 ms 20, whereas the T₂ of Pi in breast cancer tissue was 87 ± 8 ms 21. Here, it also seems likely that the short T₂ of Pi in breast cancer tissue is caused by the exchange of Pi with γ-ATP via the glycolytic pathway because cancer cells have up-regulated glycolysis.

CONCLUSION

We determined the apparent T₂s of PE, GPC, and GPE to be approximately 200 ms and that of PC to be approximately 129 ms. We attribute the lower apparent T₂ value of PC to the overlap of this resonance with the 3P and 2P resonances of 2,3-DPG from blood, with a shorter T₂. Omitting the FID signal and the first echo of PC leads to a variance-weighted T₂ of 182 ± 7 ms, whereas a biexponential fit leads to a variance-weighted T₂ for PC of 203 ± 4 ms. The short T₂ of Pi of only 86 ms is subscribed to the fast reversible exchange with γ-ATP, which is mediated by GAPDH and phosphoglycerate kinase within the glycolytic pathway.