Transverse relaxation of cerebrospinal fluid depends on glucose concentration

Abstract

Purpose:

To evaluate the biophysical processes that generate specific T₂ values and their relationship to specific cerebrospinal fluid (CSF) content.

Materials and methods:

CSF T_2s were measured ex vivo (14.1 T) from isolated CSF collected from human, rat and non-human primate. CSF T_2s were also measured in vivo at different field strength in human (3 and 7 T) and rodent (1, 4.7, 9,4 and 11.7 T) using different pulse sequences. Then, relaxivities of CSF constituents were measured, in vitro, to determine the major molecule responsible for shortening CSF T₂ (2 s) compared to saline T₂ (3 s). The impact of this major molecule on CSF T₂ was then validated in rodent, in vivo, by the simultaneous measurement of the major molecule concentration and CSF T₂.

Results:

Ex vivo CSF T₂ was about 2.0 s at 14.1 T for all species. In vivo human CSF T₂ approached ex vivo values at 3 T (2.0 s) but was significantly shorter at 7 T (0.9 s). In vivo rodent CSF T₂ decreased with increasing magnetic field and T₂ values similar to the in vitro ones were reached at 1 T (1.6 s). Glucose had the largest contribution of shortening CSF T₂ in vitro. This result was validated in rodent in vivo, showing that an acute change in CSF glucose by infusion of glucose into the blood, can be monitored via changes in CSF T₂ values.

Conclusion:

This study opens the possibility of monitoring glucose regulation of CSF at the resolution of MRI by quantitating T₂.

1. Introduction

Quantification of both longitudinal (T₁) and transverse relaxation times (T₂) has been widely used to help diagnosis and monitor the treatment of multiple diseases. Recently, there has been an increase in interest in quantitating relaxation times to provide more information. Examples include using relaxation time measurements to quantitate myelination in white matter to distinguish stages of multiple sclerosis [1], estimating iron levels during the progression of Alzheimer’s disease [2], to segment brain throughout neurodevelopment [3,4], and to monitor hippocampal damage due to epilepsy [5,6]. Although quantitating relaxation times is a powerful tool, the underlying biophysical processes that ultimately generate T₁ and T₂ values and their relationships to tissue and CSF biochemistry, remain poorly understood.

The literature reports a range of values for brain relaxation times. In healthy brain tissue, there are relatively consistent results for T₁ measured at 1.5 T and above (0.9 s − 1.4 s) depending on tissue type and magnetic field. There are also relatively consistent results for T₂ of brain tissue ranging from 0.05–0.1 s also depending on field strength and specific sequences used [7–9]. Remarkably, CSF T₂ values in the literature are very wide, ranging from 0.2 to 2.2 s (11-fold variation). Like tissue T₁ and T₂, CSF T₁ values are relatively more consistent ranging from 2.8 to 4.6 s (2-fold variation). Table 1 gives a few examples of CSF T₂ reported in the literature. In all cases, T₂ is shorter than T₁ indicating that constituents of the CSF are having a significant effect on relaxation times. There is no report that indicates which constituents of CSF explain the shorter T₂ vs T₁. The literature describes CSF composition as a dilute mix of proteins (14–40 mg/dL [10]), lipids (1.03 mg/dL; [11]), glucose (50–79 mg/dL; [12]) and metals (3 mg/dL counting all 22 metals reported in the literature [13]). There are over 2630 proteins in healthy patient CSF [10] and the three most abundant ones are albumin (20 mg/dL), IgG (2 mg/dL; [14,15]) and transferrin (2 mg/dL; [16]). Other abundant CSF proteins such as IgA, IgM, haptoglobin and α2-macroglobulin have a concentration lower than 1 mg/dL [17]. Lipids are mostly represented by phospholipids (0.4–0.6 mg/dL), cholesterol (0.3–0.5 mg/dL), glycolipids (0.07–0.09 mg/dL) and lipoproteins (ApoE =0.5 mg/dL). The most concentrated metals in CSF are Mg (3 mg/dL), Fe (0.007 mg/dL) and Zn (0.005 mg/dL) [13].

Table 1.

CSF relaxation times T₁ and T₂ from different reports in the literature. Relaxation times were measured in vivo in healthy human at different field strengths.

Magnetic field (T)	Human CSF in vivo		Literature
	T1 (s)	T 2 (s)
0.15	4.36	1.76	Hopkins et al., 1986 [32]
0.35	2.72	0.17	Kjos et al., 1985 [40]
0.6	4.22	2.19	Hopkins et al., 1986 [32]
1.4	4.31	-	Hopkins et al., 1986 [32]
1.5	-	0.15	Larsson et al., 1989 [33]
3	-	0.50	Piechnik et al., 2009 [29]
4	4.55	0.70	Jezzard et al., 1996 [30]

The purpose of the present work was to determine the T₂ of CSF ex vivo and in vivo and explore the possible reasons for such a large discrepancy of T₂ values in the literature. Furthermore, it was determined which constituents of CSF are involved in generating these specific T₂ values.

2. Materials and methods

The Institutional Review Board approved all human study protocols, and informed consent was received from all participants. MRI and CSF studies were performed on human subjects recruited for natural history studies of multiple sclerosis and HTLV-1 Associated Myelopathy/tropical spastic paraparesis (HAM/TSP). All animal experiments were performed in accordance with NIH guidelines and were approved by the Animal Care and Use Committee of the National Institute of Neurological Disorder and Stroke, National Institutes of Health (Bethesda, MD USA).

2.1. Ex vivo and in vitro MRI experiments

2.1.1. Ex vivo CSF samples

CSF samples were collected from 3 different species: 3 healthy human subjects (all males, age of 36.7 ± 10 years), 3 rhesus macaques (2 males, 1 female, mean age 9 ± 4 years), and 4 CD rats with a permanent intracisternal cannulation (all males, 3 months old; Charles River, USA). CSF was collected by lumbar puncture in humans in the context of ongoing clinical studies at the NINDS, NIH. CSF was obtained from non-human primates by transcutaneous cisterna magna puncture (one-time collection, 1–2 mL of CSF) and CSF from rats was obtained via a daily CSF collection (a maximum of 100 μL of CSF/collection was allowed) using the permanent intracisternal cannulation until reaching a final volume of 500 μL. Sampling periods was 24 h apart to allow for CSF production.

2.1.2. Standard in vitro solutions for T₂ comparisons

To correlate T₂ values with metal, protein and glucose concentration, the following solutions were prepared in 0.9% NaCl: bovine serum albumin (1.5–30 μM; BSA, Bio Rad, USA), ZnCl₂ (0–1 mM; Sigma-Aldrich, USA), FeCl₂ (0–1 mM; Sigma-Aldrich, USA), MnCl₂ (0–80 μM; Sigma-Aldrich, USA), CuCl₂ (0–1 mM, Sigma-Aldrich, USA) and glucose (0– 1 mM; Sigma-Aldrich, USA). BSA was used to mimic protein content of CSF. The pH range of these solutions was between 6.2 and 7.4.

To evaluate whether chemical exchange due to pH explains T₂ differences at different field strength, glucose and BSA solutions were prepared in 0.9% NaCl at a physiological concentration of 4 mM and 0.01 mM, respectively (see Fig. 4). The pH was adjusted to 2 by adding few drops of HCl 50 mM and to 10 by adding few drops of NaOH 50 mM.

Fig. 4.

CSF compounds that can substantially change human CSF T₂ relaxivity. All T₂ measurements were performed on a 14.1 T Bruker MRI system at 37 °C using a CPMG sequence with TE(2048) = 2/2048 ms (intervals of 1 ms). A. Example of ΔR₂ plotted against glucose (n = 3) and BSA concentration (n = 3). The regression equation is written for each graph, showing the relaxivity r₂ for each species represented in this figure. B. Table showing the relaxivity r₂, the concentration and the calculated R₂ in human CSF for each element. Relaxivity r₂ was computed for each species (proteins, metals and glucose) from the plot between ΔR₂ and solute concentration. The concentration of metals in human CSF was determined by IC-PMS and the one of BSA and glucose was quantified using colorimetric methods. Using r₂ and concentration values, we calculated the R₂ for each element in human CSF (example for the BSA: R₂ = r₂ × [BSA concentration]; R₂ = 7.6 × 8.9.10⁻³; R₂ = 6.8.10⁻² s⁻¹).

2.1.3. Ex vivo and in vitro MRI acquisition

All ex vivo and in vitro experiments were performed in a vertical bore, 14.1 T MRI system (Bruker Biospin, Billerica, MA) using a volume transmit/receive coil. Ex vivo CSF and in vitro BSA and glucose solutions were also measured at 4.7 T MRI using a volume transmit and receive coil. In order to maintain constant temperature of the samples, the internal MRI temperature was set at 37 °C. Solutions (CSF or solute) were transferred in a 1 cm NMR glass tube and placed in a 37 °C water bath for at least 20 min prior to each experiment. Each tube was homogenized by vortex and directly placed in the NMR coil. A 3D image was performed to determine the sample position in the MRI. T₂ measurement of the entire sample was performed with a multiecho Car-Purcell-Meiboom-Gill (CPMG) spectroscopy sequence with TR = 20,000 ms, TE(2048) = 2/2048 ms (intervals of 1 ms) and time of acquisition (TA) = 2 min 40 s. This eliminated the presence of any residual imaging gradients. T₁ mapping was performed for ex vivo CSF samples only with a 2D RARE-VTR sequence with TR(6) = 500–20,000 ms; TE = 7.6 ms; TA = 9 min 25 and resolution of 0.625 × 0.625 mm.

2.2. In vivo MRI experiments

2.2.1. Human

In vivo CSF T₂ values were calculated from three healthy subjects (1 male and 2 females, mean age 39 ± 4 years).

A heavily T₂-weighted image was performed to visualize CSF without any contribution of brain tissue in the image. Conventional T₂ maps were obtained with a standard two point T₂ sequence and T₂ maps were obtained with a multiple TEs sequence that attempted to minimize applied field gradients. All these sequences were used on 2 healthy subjects (2 females, ages of 42 and 35 years) on a Siemens 3 T Skyra system (software version VD11) equipped with a 32-channel phased-array receive-only head coil. Conventional T₂ maps were generated using a dual-echo fast-spin-echo sequence with TR = 3500 ms, TE = 11/101 ms, TA = 3.5 min, in-plane resolution 0.5 mm, and slice thickness 3 mm. The T₂ mapping was a 2D spin-echo sequence (Siemens product sequence) with TR = 10,000 ms and 32 TEs equally spaced between 30.5 and 945.5 ms, TA = 10.5 min, in-plane resolution 3 mm, and slice thickness 3 mm. Heavily T₂-weighted images were acquired using a three dimension, fast spin–echo sequence (FSE), prescan-normalize filter for receive coil homogeneity, TR = 4800 ms, TE = 752 ms, FA = 100°, scan acceleration factor GRAPPA = 2, echo-train-length = 421, fat-saturation mode = strong, 0.65 mm isotropic resolution, bandwidth = 543 Hz/Px, and TA = 4 min 43 s [18]. This last sequence was also used on a Siemens 7 T MRI on one healthy volunteer (40 y.o. male) with the same parameters except the resolution: in-plane resolution 1.5 mm, and slice thickness 1.5 mm.

2.2.2. Rodent

A total of fifteen rodents were used to perform heavily T₂-weighted images, conventional T₂ mapping (n = 3), and T₂-spectroscopy (n = 12) to remove the presence of any background imaging sequence gradientsFor all the scans, rodents were anesthetized with isoflurane (4% induction, 1.5% maintenance) and placed in a dedicated cradle equipped with byte and ear bars. Temperature and respiratory rate were monitored during the acquisition. A 3D image was performed to determine the animal position in the MRI.

2.2.2.1. Conventional T₂ mapping and heavily T₂ images.

Three rats (SD, 3 months old) were used for this experiment. Heavily T₂-weighted images and conventional T₂ mapping were performed on a 4.7 T MRI system (Bruker Avance III scanner and Bruker console) using a 72 mm ID birdcage coil and a surface coil to detect.

Three dimensional, heavily T₂-weighted images were acquired with a T₂-preparation (128 echoes, 3 ms spacing, M-LEV8 phase cycling) with a segmented FLASH readout [19], TR = 10 ms, TE = 384 ms, in-plane resolution 0.3 mm, and slice thickness 1 mm, TA = 24 min. Conventional T₂ maps were generated with a multi-slice-multi-echo (MSME) sequence, TR = 4000 ms, TE(20) = 10.95/219 ms (intervals of 10.95 ms), in-plane resolution 0.3 mm, slice thickness 1 mm and TA = 6 min 24 s. For the MSME sequence, we used crushers giving 2π phase dispersion across the slice (same gradient strength for all echoes), and a 1-ms windowed sinc pulse for the refocusing pulse.

2.2.2.2. T₂-spectroscopy.

Twelve mice (C57Bl6 male, 3 months old) were used to performed T₂-spectroscopy. T₂ quantitation was carried out using T₂ spectroscopy sequence on 4 different horizontal field strength MRIs: 1 T (Bruker, Icon), 4.7 T (Bruker MRI), 9.4 T (Bruker MRI) and 11.7 T (Bruker MRI). The experiments at 4.7, 9.4 and 11.7 T all used a large birdcage coil (72 mm I.d. at 4.7 and 89 mm at 9.4 and 11.7 T) for transmission and a 1 cm diameter surface coil for reception. This arrangement serves to localize the experiment, and in particular, reduces the signal from the aqueous part of the eye (to b 10% of the signal).

The spectroscopic evaluation of T₂ was obtained using a multiecho CPMG sequence with TR = 10,000 ms, TE(1024) = 2/2048 ms (intervals of 2 ms), Nav = 8, and TA = 1 min 20 s at 1 T; TR = 20,000 ms, TE(1024) = 2/2048 ms (intervals of 2 ms), Nav = 8 and TA = 2 min 40 s at 4.7 T, 9.4 T and 11.7 T. This spectroscopic evaluation of T₂ was performed in vivo (CSF flow) and post mortem (no CSF flow) to observe the eventual impact of CSF flow on CSF T₂ value. At the end of each in vivo scan, the mouse was euthanized by cerebral dislocation and immediately re-scanned to obtain the post mortem CSF T₂ value. Due to the short time between sacrificing the animal and MRI data collection it is not likely that the pH got acidic due to the long time scale of CSF acidification [20].

2.3. In vivo experiments to manipulate CSF glucose concentrations: glucose challenge

To determine if T₂ of CSF in vivo was sensitive to changes in glucose, experiments were done to manipulate CSF glucose in rodent. It is well established that CSF glucose changes with blood glucose concentrations [21], consequently, glucose was injected in the blood. Ten rats (SD, male, 3 months old) were used for this study. Five rats were fasted for 24 h before the experiment, and five other rats were fed ad libitum. The day of the experiment, rats were anesthetized with isoflurane (4% induction, 1.5% maintenance), and a non-survival surgery was performed to catheterize the femoral artery for blood collection and the jugular vein for glucose infusion. Two PE50 tubes of about 1 m each were used for the catheterization and blood clotting was avoided using 10% heparin (Hospira, USA) in the saline solution. Dead volume was about 0.4 mL. After surgery, the rat was placed in a dedicated cradle equipped with bite and ear bars. Temperature and respiratory rate were monitored during the acquisition.

Experiments were performed at 4.7 T (Bruker MRI), using a volume excitation and surface coil detection configuration. A 3D image was performed to determine the animal position in the MRI. T₂ measurements of the whole head was performed using a multi-echo CPMG sequence with TR = 20,000 ms, TE(1024) = 2/2048 ms (intervals of 2 ms). This measurement was repeated 3 times, every 30 min, for a total time of 1.5h. Before each T₂ measurement, glycemia was measured using a glucometer (OneTouch UtraMini, Lifescan, USA) and one-time usage test strips (OneTouch Utra, Lifescan, USA). For this measurement, blood was collecting from the femoral vein catheter going out of the MRI scanner. At the end of the first T₂ measurement, a bolus of glucose (0.4 mL, 20% glucose in water) was injected via the jugular vein catheter, followed by continuous glucose infusion (10% in water, 0.05 mL/min). The infusion was stopped at the end of the experiment, just after the end of the third T₂ measurement. The rat was then placed in a stereotaxic frame and the skin of the neck was shaved and opened. The cisterna magna was exposed by separating the neck muscles and then gently perforating with a 25G needle. CSF was then directly collected using a glass capillary and rapidly frozen at −20 °C for glucose analysis (see below).

2.4. MRI data analysis

2.4.1. Human

T₂ maps were calculated by performing a bi-exponential fit of the pixel intensities using the robust least absolute residuals method in Matlab (MathWorks, MA). CSF-ROIs were drawn by hand in the lateral ventricles. The median T₂ value within the CSF-ROI was reported as the CSF-T₂ value for each subject.

2.4.2. Rodent

T₂ mapping: T₂ values were calculated by performing a mono-exponential fit of the pixel or echo intensities. ROIs were drawn on the lateral ventricles, and the T₂ maps were derived by fitting.

T₂ measurement by spectroscopy: The T₂ data was fitted using a linear least squares with nonnegativity constraints method (NNLS; Matlab lsqnonneg function, MathWorks; [22]). For the data presented here, no regularization was used. T₂ values were assigned to 1000 bins with an equal separation of 2 ms (covers 2–2000 ms) for the fitting. We obtained a spectrum with three groups of peaks. The first group was assigned to brain and muscle tissue T₂, the second was assigned to fat T₂ and the third was assigned to CSF T₂ (Fig. 2B). The average value of peak assigned to CSF was used to quantitate the T₂.

Fig. 2.

In vivo human and rodent CSF T₂ measurement at different field strength. A. CSF T₂ was measured at different field strengths. In humans, T₂ mapping was performed at 7 T and 3 T. Images show an apparently longer CSF T₂ at 3 T than at 7 T (according to the color scale). The blue boxes represent the ROIs drown in the lateral ventricles. B. In rodent, spectroscopic sequences were used to measure CSF T₂ of the whole head at 11.7 T, 9.4 T, 4.7 T, and 1 T. There are 3 peaks in the spectra, the first corresponding to parenchyma with very short T₂ (0–0.1 s), the second to fat with short T₂ (0.1–0.2 s), and the last to CSF with the longest T₂ (0.3–2 s). C. Quantification of CSF T₂ in vivo shows longer T₂ values at low field in both Human and rodent. For rodent, CSF T₂ post mortem got longer at higher field likely due to CSF flow in the brain/head gradients.

2.4.3. Solutions (CSF and solute)

T₂ was extracted from spectroscopic measurements using NNLS method [23]. Only a single component was found in the solution data and the average value of this component was used.

For each solute (proteins, metals and glucose), ΔR₂ (1/T₂; s⁻¹) values were plotted against solute concentration (mM). Relaxivity r₂ (mM⁻¹·s⁻¹) was then computed (ΔR₂ = r₂ × [solute]). Each solute’s concentration in healthy human CSF was measured. Based on this concentration, ΔR₂ value in human healthy CSF for each solute was calculated using ΔR₂ = r₂ × [BSA] + b assuming the in vitro values for relaxivities.

CSF protein concentration was quantified according to the method of Bradford (Quick start Bradford protein kit, Bio Rad, USA). A reference of 1 mg·mL⁻¹ of BSA was used [24]. CSF paramagnetic metal (Cu, Fe, Mn, and Zn) concentration was quantified by inductively coupled plasma mass spectrometry technique (IC-PMS, Exova, California). CSF glucose concentration was performed using a colorimetric technique (Gluc kit, Sigma, USA). A reference of 1 mg·mL⁻¹ of glucose was used. T₁ mapping: T₁ values were calculated by performing a mono-exponential fit of echo intensities. ROIs were manually drawn in the area cor-responding to liquid in the tube and T₁ map was derived by fitting.

3. Results

3.1. Ex vivo CSF T₂ measurement

Data from human, rodent and non-human primate (NHP) are shown in Table 2. Saline was used as a control and had a T₂ of about 3.0 s. The average human CSF T₂ was 2.1 s. Rodent and NHP T₂s were 1.8 s and 1.9 s, respectively. Representative signal T₂ decay is shown in Sup Fig. 1.

Table 2.

T₁ and T₂ of ex vivo CSF samples. CSF samples were collected from healthy humans, rats and monkeys. The relaxation times were measured at 37 °C on a 14.1 T Bruker scanner using a multi-echo CPMG sequence with TR = 20,000 ms, TE(2048) = 2/2048 ms (intervals of 1 ms). Values are represented as mean ± standard deviation.

Sample name	n	T1 (s) at 14.1 T	T2 (s) at
Saline	4	4.20 ± 0.29	3.07 ± 0.24
Human	3	4.22 ± 0.30	2.14 ± 0.11
Rodent	4	3.98 ± 0.03	1.79 ± 0.04
NHP	3	4.04 ± 0.21	1.87 ± 0.13

3.2. CSF T₂ values from MRI are sequence and magnetic field dependent

Heavily T₂-weighted images of human brain (Fig. 1A) and rodent brain (Fig. 1C) show high signal within the CSF surrounding the brain and inside the ventricles. The contrast to noise ratio between CSF and tissue was ~ 30:1 for human and ~ 16:1 for rodent. A conventional T₂ mapping measured a T₂ of 887 ± 50 ms for human CSF at 3 T and 174 ± 27 ms for rodent CSF at 4.7 T (Fig. 1B and D). Representative T₂ decay curve for human and signal T₂ decay for rodent are shown in Sup Figs 2 and 3, respectively.

Fig. 1.

In vivo human and rodent CSF T₂ measurement. Human imaging was performed on a 3 T Siemens Skyra and rodent imaging on a 4.7 T Bruker. Heavily T₂-w images are shown in panel A (TE = 752 ms) and C (TE = 384 ms). B. Human T₂ mapping was performed using a FSE sequence with TR = 3500 ms, TE(2) = 11/101 ms, in-plane resolution 0.5 mm, and slice thickness 3 mm. D. Rodent T₂ mapping was performed using a MSME sequence, TR = 4000 ms, TE(20) = 10.95/219 ms (intervals of 10.95 ms), in-plane resolution 0.3 mm, and slice thickness 1 mm. ROIs were manually drawn in the lateral ventricles and the resulting T₂ values were 887 ± 50 ms for human CSF and 174 ± 27 for rodent CSF.

Changes in the T₂ mapping sequence to minimize applied imaging gradients led to an increased CSF T₂ measured from human brain of 2.02 s at 3 T (Fig. 2A). This is consistent with the ex vivo T₂ measurement at low field (4.7 T) and indicates that judicious use of MRI sequence parameters can get an accurate T₂ at 3 T. Interestingly, the apparent human CSF T₂ dropped to 850 ms at 7 T even though the sequence was the same as that used at 3 T. Representative T₂ curves are shown in Sup Fig. 2.

To explore the reasons for the apparent shorter T₂ at high field, the rodent brain was studied. To eliminate the possibility of MRI sequence causing effects, a spectroscopic T₂ measurement was made from the entire rodent head (Fig. 2B). The long T₂ of CSF compared to fat or tissue T₂ enabled an unambiguous assignment of CSF in the T₂ relaxograms. An apparent T₂ of 0.4 s was measured at 11.7 T significantly shorter than the ex vivo measurement. CSF T₂ was measured at 9.4, 4.7, and 1 T to determine the effects of magnetic field. The T₂ of rodent CSF in vivo was 1.5 s at 1 T approaching the ex vivo measurement of 1.8 s. Representative signal T₂ decay is shown in Sup Fig. 4.

CSF flow in and out of the slice or through residual field gradients would be expected to shorten CSF T₂. To test this hypothesis, CSF T₂ was measured in live rodent (CSF flowing) and in post mortem rodent (no CSF flow). CSF T₂ values were about the same in live and post mortem rodent at 1 T where residual magnetic field gradients are small (1.6 s; Fig. 2C). There was an increasing difference as the field strength increased and at 11.7 T there was a 77% difference in T₂ from live (0.4 s) and post-mortem rodent (0.7 s; Fig. 2C).

Data for the magnetic field dependence of ex vivo and in vivo rodent and human CSF R₂ as well as saline R₂ is plotted in Fig. 3. There is almost no change of saline R₂ measured at 4.7 T and 14.1 T (ratio of 0.94) and a bigger change was observed for ex vivo CSF of human and rodent between 4.7 T and 14.1 T (ratio of 0.85 and 0.70, respectively). Both in vivo human and rodent CSF R₂s decrease with decreasing field strength. The in vivo human CSF R₂ approaches the ex vivo values by 3 T and the in vivo rodent CSF R₂ approaches the ex vivo values only at 1 T.

Fig. 3.

R₂ differences between saline, ex vivo and in vivo rodent and Human CSF at different field. A. Rodent CSF and saline T₂ measurement were performed using a multi-echo CPMG sequence with TR = 10,000 ms at 1 T and TR = 20,000 ms at 4.7 T, 9.4 T and 14.1 T; TE(1024) = 2/2048 ms (intervals of 2 ms) at 1, 4.7 and 9.4 T and TE(2048) = 2/2048 ms (intervals of 1 ms) at 14.1 T. Scans were performed at 1, 4.7. 9.4 and 11.7 T for in vivo data and at 14.1 and 4.7 T for ex vivo data. B. Human CSF T₂ measurements were performed using two different sequences. In vivo CSF T₂ measurements were performed by a multi-contrast spin echo sequence with TR = 10,000 ms and 32 TEs equally spaced between 30.5 and 945.5 ms. Ex vivo CSF and saline T₂ measurements were performed using the same multi-echo CPMG sequence detailed above for rodent.

3.3. What causes the shortening of CSF T₂ as compared to saline T₂?

To test whether paramagnetic metals, ions, proteins, and/or glucose can explain the T₂ relaxivity of CSF as compared to saline, in vitro relaxivities were measured. BSA was used as a surrogate for CSF protein. Relaxivity data for Zn, Cu, Fe, Mn, glucose and BSA is shown in Fig. 4. The concentration of each of the relevant species was measured in CSF samples (Fig. 4B). The concentration of metals was too low to significantly modify CSF T₂ relaxivity (b 0.5%). Proteins can explain approximately 15% of the shortening of CSF T₂. Interestingly, the glucose concentration is high enough to cause most of the shortening to CSF T₂ with respect to saline T₂. Indeed, the human CSF R₂ of approximately 0.5 is explained primarily by the R₂ of saline (approximately 0.3) plus the R₂ of glucose (approximately 0.2). The T₂ relaxivity of glucose is quite low (0.067 mM⁻² s⁻¹) but the very long T₂ of CSF enables the high concentrations of glucose (~ 3 mM) to dominate the T₂ relaxivity of CSF.

3.4. Chemical exchange does not explain the in vivo vs ex vivo T₂ discrepancy at high field

A major mechanism to explain relaxation by glucose is exchange of protons between glucose and water. To test whether chemical exchange can explain the difference of T₂ values at high field between in vivo and ex vivo (data shown in Fig. 3), in vitro relaxivities of glucose and BSA solutions at different pH were measured. These data were compared to in vivo and ex vivo rodent data, as well as saline and shown in Fig. 5. The range of glucose solutions R_2s at pH 2, 7 and 10 were between 0.27 ± 0.02 and 0.65 ± 0.03 at 14.1 T and between 0.36 ± 0.01 and 0.59 ± 0.02 at 4.7 T. For the BSA, the range of R₂ at pH 2, 7 and 10 were between 0.26 ± 0.02 and 0.39 ± 0.04 at 14.1 T and between 0.34 ± 0.03 and 0.41 ± 0.01 at 4.7 T. These data overlap with ex vivo CSF R₂ (0.56 ± 0.01 at 14.1 T and 0.39 ± 0.03 at 4.7 T). Glucose R₂, at different pH are very different from in vivo CSF R₂ at high field (2.6± 0.1 at 11.7 T) suggesting that chemical exchange is not a major contributor to the T₂ difference between in vivo and ex vivo data at high field.

Fig. 5.

Effect of chemical exchange on rodent CSF R₂ values. T₂ measurement of rodent CSF in vivo and ex vivo, saline, glucose and BSA solutions were performed using a multi-echo CPMG sequence with TR = 10,000 ms at 1 T and TR = 20,000 ms at 4.7 T, 9.4 T and 14.1 T; TE(1024) = 2/2048 ms (intervals of 2 ms) at 1, 4.7 and 9.4 T and TE(2048) = 2/2048 ms (intervals of 1 ms) at 14.1 T. Scans were performed at 1, 4.7. 9.4 and 11.7 T for in vivo data and at 14.1 and 4.7 T for ex vivo data as well as for saline, glucose and BSA solutions. Glucose and BSA solutions were prepared in saline at a concentration of 4 mM and 0.01 mM, respectively, to mimic CSF physiological values. The pH of both, glucose and BSA solutions is at 7 and it was changed to be acidic (pH 2) and basic (pH 10).

3.5. CSF T₂ depends on serum glucose concentration

The strong contribution of glucose to the CSF relaxivity opens the possibility of measuring changes in CSF glucose by measuring changes in CSF T₂. An in vivo experiment was performed to determine if CSF T₂ changes with changes in CSF glucose concentration. To increase the size of the glucose change caused by infusion, five animals were fasted to lower blood and CSF glucose concentrations and five other animals were fed ad libitum. One of the “fed ad libitum” rat could not be used, consequently the total of rats was nine. Experiments would have been best done at 1 T where the measured T₂ is longest (best dynamic range), however, due to availability of MRI time, these experiments were performed at 4.7 T. For each rat, CSF puncture was performed at the end of the experiment, before euthanasia; because of the technical difficulty of CSF puncture only seven CSF samples were obtained from the nine rats that were used in this experiment. The protocol used and the R₂ results are show in Fig. 5. There was good correlation between the R₂ measured in both the blood glucose (Fig. 5B) and CSF glucose (Fig. 6C). The correlation between CSF T₂ and blood glucose (glycemia) had a R₂ = 0.71 (p b 0.001; Fig. B) and the correlation between CSF T₂ and CSF glucose (glycorrhachia) had a R₂ = 0.93 (p b 0.005; Fig. 6C). Thus, CSF T₂ values change when CSF glucose is varied in this model of increased blood glucose.

Fig. 6.

CSF T₂ depends on glycemia A. Experimental scheme. Surgery was performed under anesthesia to catheterize both jugular vein and femoral artery. Black arrow corresponds to serum glucose concentration measurement via the femoral artery. The rat is kept within the MRI scanner (4.7 T Bruker MRI system). B. CSF R₂ values measured at 3 time point as shown in A using the spectroscopy sequence, plotted against serum glucose concentration values (n = 9 rats). C. CSF R₂ values measured at the end of the experiment, just before CSF collection, plotted against CSF glucose concentration values (n = 7). CSF was collected just after the last MRI acquisition and before euthanasia. D. Glycorrhachia as a function of glycemia (n = 7 rats). The ratio of glycorrhachia to glycemia is 0.42.

The ratio of rodent CSF glucose and serum glucose was 0.41 (Fig. 6D) which is close to the value of 0.6 previously measured in rodent and human [21,25]. The apparent T₂ relaxivity for glucose derived from this experiment was 0.13 mM⁻¹ s⁻¹ (slope of line in Fig. 6C), higher than the in vitro value of 0.07 mM⁻¹ s⁻¹. These discrepancies are likely due to the fact that the protocol was not designed to make sure glucose concentrations in the CSF had fully equilibrated and were sampled at the same time as T₂ was measured.

4. Discussion

In this study, the transverse relaxation time (T₂) of CSF was measured ex vivo and in vivo to determine the values for CSF T₂ and understand the factors that may lead to N 10-fold variation in CSF T₂s in the literature (Table 1). In vitro CSF T₂ measurements were approximately 2 s for human, non-human primate and rodent. A similar value for T₂ of CSF could be obtained from human brain at 3 T and from rodent brain at 1 T. Higher field measurements led to apparent shorter T₂ most likely due to residual field gradients in the CSF caused by the brain tissue at the higher fields. While the ex vivo T₂ value of CSF was long, it was still shorter than the T₂ of saline. Glucose was identified as the predominant CSF component which can explain the shorter CSF T₂ in vitro as compared to saline. This opens the possibility of using CSF T₂ to monitor CSF glucose. This was confirmed in vivo in rodent by glucose infusion and simultaneous CSF T₂ measurement, opening the possibility of studying glucose regulation of CSF at MRI resolution.

4.1. In vivo CSF T₂ depends on field strength and sequences parameters

Ex vivo, CSF T₂ was much longer than a conventional in vivo CSF T₂ measurements using standard FSE or MSME sequences both in human and rodent brain. This result agrees with previous concerns raised in the literature. Indeed, it has been shown that conventional T₂ mapping sequences could be affected by residual field gradients and spurious echoes from these sequences [26,27]. Therefore, minimizing the residual imaging gradients using an adequate sequence and decreasing residual tissue gradients by using lower magnetic field strength as we did in our study, led to longer apparent T₂ measurements of CSF. Indeed, human in vivo CSF T₂ approached the in vitro data at 3 T while the rodent in vivo CSF T₂ approached in vitro data at 1 T. Longer echo times should lead to shorter apparent T₂s due to longer diffusion distances for water in the residual gradients. Flow of CSF in residual gradients could also shorten T₂ [28]. Therefore, in our study we showed that eliminating CSF flow in the post-mortem rodent brain lengthened the measured T₂ excepted at 1 T, where residual tissue gradients had a minimal effect on T₂.

A spectroscopy pulse sequence was used in rodents to eliminate imaging gradients completely which achieved the longest T₂’s measured here. However, even at 1 T rodent CSF T₂ values in vivo did not reach the long values of T₂ measured ex vivo implying that tissue gradients still contribute to the shortening of T₂. Also, tissue gradient are expected to have larger impact on rodents CSF T₂ due to the small size of the CSF spaces compare to that in humans. In the human brain, a simple multiple TE spin echo acquisition with a long TR led to the longest T₂s as compared to clinically conventional used two point T₂ mapping pulse sequence based on a fast spin echo. At 3 T, human CSF measured was about 2 s which is on the low end of values obtained ex vivo (2–2.8 s).

Implementing a spectroscopic measurement of human CSF T_2, as was done for the rodents, would allow us to assess how many residual imaging gradients decreased the T₂ at 3 T. At 7 T, it is clear that tissue gradients begin to shorten the apparent CSF T₂ in humans since the CSF T₂ dropped from 2 to 0.85 s even when using the same MRI pulse sequence as used at 3 T.

Previous data from both, human and rodent, show a large range of values for T₂. Typically, these measurements did not attempt to minimize residual gradients that might affect T₂ measurements [28,29,30]. One report has longer CSF T₂ values [30,31,32,33] which were measured at low field and likely used imaging pulse sequences with lower applied field gradients available at the time. In vitro human CSF T₂ values have been reported to be 2.64 s at 2.33 T [34] and 1.82 at 0.25 T [35] consistent with the results in the present study.

4.2. CSF T₂ depends on glucose concentration

Even though CSF has a very low content of biological molecules, the ex vivo T₂ of CSF in this study was still shorter than the T₂ of saline. In contrast, CSF T₁ was very similar to saline T₁. It is well known that relaxation agents tend to have a higher T₂ relaxivity than T₁ relaxivity. To understand the origin of the shorter T₂ of CSF as compared to saline, the relaxivity of major constituents of CSF were measured and compared to measured concentrations in CSF. The results indicate that metals concentrations were too low to significantly change the CSF T₂ while protein concentration was high enough to have a small effect on CSF T₂. In this study, we used albumin relaxivity as a representative of CSF protein. Thus, differences in protein relaxivities will probably affect this determination to some extent. Glucose is the component with the highest concentration in CSF and interestingly, the T₂ relaxivity of glucose is large enough to affect CSF T₂ proportionally to glucose CSF concentration. These results suggest that CSF T₂ can be used as a measure for CSF glucose concentration. Consistent with this, a change in CSF T₂ was measured as CSF glucose was varied in the rodent brain by varying the blood glucose level with infusion of glucose. This opens the possibility of studying regulation of glucose levels in human CSF at the MRI resolution.

There has been recent interest in using glucose as a contrast agent for MRI. Yadav et al., showed that T₂ of blood depends on glucose concentration [36]. They found a glucose T₂ relaxivity of 0.02, 0.06, and 0.08 mM⁻¹ s⁻¹ at 3.0, 7.0, and 11.7 T, respectively. The relaxivity in blood at 11.7 T was 0.09 mM⁻¹ s⁻¹. In agreement with their results, a relaxivity of glucose in water of 0.07 mM⁻¹ s⁻¹ at 14.1 T and 0.04mM⁻¹ s⁻¹ at 4.7 T was measured. There is a field dependence for all solutions most likely due to chemical exchange effects. For CSF, we show that exchange can be important for explaining the difference between in vivo and in vitro T₂ relaxivity but this is not large enough to explain differences at high field where the background gradient effects most likely explain the difference between in vivo and in vitro T₂. However, for accurate quantitation of glucose from T₂, care will have to be taken to account for factors that may affect exchange such as CSF pH. Yadav et al. performed an in vivo study after a glucose bolus injection while imaging the liver, and demonstrated a 10% drop in signal intensity after glucose infusion followed by recovery of the signal intensity after about 50–100 s. They concluded that glucose can be used as a T₂ contrast agent for MRI at concentrations that are already approved for human use.

Another approach to detect glucose as a possible MRI contrast agent is to measure chemical exchange of protons from glucose to water using the chemical exchange saturation transfer (CEST) MRI. Glucose, used as a CEST agent (glucoCEST), was first described by two groups: Walker-Samuel et al. and Chan et al. and shown to be detectable in rodent brain [37,38]. Recent work by Xu et al., used natural D-glucose as a contrast agent for CEST [39]. They explored the feasibility of using D-glucose for dynamic perfusion imaging to detect malignant mouse brain tumors based on blood-brain-barrier breakdown after an intravenous bolus of D-glucose. The time-resolved glucose signal changes were detected using glucoCEST MRI. They showed that dynamic glucoCEST MRI is a feasible technique for studying brain tumor enhancement, reflecting differences in tumor blood volume and permeability with respect to normal brain. The long T₁ of CSF allows glucoCEST to be performed in CSF. In general, CEST experiments are much less sensitive than T₂ weighted MRI, however, they do not suffer from the issues of residual gradients that affect CSF T₂ measurements. CEST studies to determine glucose may complement CSF T₂ changes especially in cases where increased protein due to tissue degradation or increased cellularity may also affect CSF T₂.

In conclusion, this work demonstrates that quantitating CSF T₂ is very difficult due to the long T₂ of CSF. Any residual gradients either from the imaging sequence or residual tissue gradients especially at higher magnetic fields leads to an apparent shortening of the measured T₂. Interestingly, both the very long T₂ of CSF and the fact that glucose is the dominant contributor to the relaxation properties of CSF lead to the intriguing possibility that regulation of CSF glucose can be studied by simply using T₂ weighted MRI.

Abbreviations:

CNR

contrast to noise ratio

CPMG

multiecho Car-Purcell-Meiboom-Gill

CSF

cerebrospinal fluid

T₁

longitudinal relaxation time

T₂

transversal relaxation time