- Email: [email protected]
MRI characteristics of the glia limitans externa: A 7T study
Magnetic Resonance Imaging 44 (2017) 140–145
Contents lists available at ScienceDirect
Magnetic Resonance Imaging journal homepage: www.elsevier.com/locate/mri
MRI characteristics of the glia limitans externa: A 7T study a
a
a
a
MARK a
Kiyotaka Suzuki , Kenichi Yamada , Kazunori Nakada , Yuji Suzuki , Masaki Watanabe , Ingrid L. Kweea,b, Tsutomu Nakadaa,b,⁎,1 a b
Center for Integrated Human Brain Science, Brain Research Institute, University of Niigata, Niigata, Japan Department of Neurology, University of California, Davis, USA
A R T I C L E I N F O
A B S T R A C T
Keywords: FLAIR ICA k-means clustering AQP-4
Purpose: To perform a systematic analysis of the intrinsic contrast parameters of the FLAIR hyperintense rim (FHR), a thin layer of high intensity covering the entire surface of the cerebral cortex detected on fluid-attenuated inversion recovery (FLAIR) sequence T2 weighted imaging performed on a 7T system, in an attempt to identify its anatomical correlate. Methods: Fast spin echo inversion recovery (FSE-IR) and cardiac-gated fast spin echo (FSE) images were obtained with defined parameters in eight normal volunteers on a 7 T MRI system to determine T2 and proton density, T1 characteristics. K-means clustering analysis of parameter sets was performed using MATLAB version R2015b for the purpose of identifying the cluster reflecting FHR. The results were subsequently confirmed by independent component analysis (ICA) based on T1 behavior on FSE-IR using a MATLAB script of FastICA algorithm. Results: The structure giving rise to FHR was found to have a unique combination of intrinsic contrast parameters of low proton density, long T2, and disproportionally short T1. The findings are in strong agreement with the functional and structural specifics of the glia limitans externa (GLE), a structure composed of snuggled endfeet of astrocytes containing abundant aquaporin-4 (AQP-4), the main water channel of the brain. Conclusion: Intrinsic contrast parameters of FHR reflect structural and functional specifics of the GLE, and their values are highly dependent on the physiologic functionality of AQP-4. Microscopic imaging on a 7T system and analysis of GLE contrast parameters can be developed into a method for evaluating AQP-4 functionality.
1. Introduction Fluid-attenuated inversion recovery (FLAIR) sequence with T2 (transverse or spin-spin relaxation time) weighting, T2-FLAIR, is a widely used magnetic resonance imaging (MRI) sequence for detecting structural brain abnormalities. The sequence is designed to suppress background high intensity of cerebrospinal fluid (CSF) on T2 weighted images by optimizing inversion time (TI) to the zero-crossing point of the longitudinal magnetization of CSF. T2-FLAIR imaging has made possible the detection of subtle abnormalities of structures adjacent to the cerebral cortex [1]. Detection of high intensity lesions on the surface of the brain and within the CSF space by T2-FLAIR on conventional MRI systems have been thought to reflect pathological conditions or artifacts [2]. MRI technology has recently entered the high and ultra-high field era with higher resolution images on 3 to 7 Tesla systems, and has led to new findings of the human cortex [3–5]. One unexpected finding is the
⁎
1
high intensities in a thin layer of normal cortex identified with T2FLAIR imaging on a 7T system. The unusual phenomenon has been referred to as “FLAIR hyperintense rim (FHR)” [5]. It is believed that FHR reflects a normal anatomical structure on the surface of the cerebral cortex. This structure possesses significantly long T2, but it is nevertheless distinct from CSF. Thus far, there have been no published studies in the English literature on the systematic evaluation of the structure corresponding to FHR. The surface of the cerebral cortex adjacent to the CSF filled subarachnoid space is enveloped by the pia matter, a thin fibrous tissue that is impermeable to fluid. This simple structure has no known properties that would lead to the unique MRI characteristic responsible for FHR. Immediately beneath the pia matter, however, is a ribbon like structure, known as the glia limitans externa (GLE), which has several unique anatomical characteristics. First, the GLE is composed of snuggled endfeet of astrocytes. Second, astrocyte endfeet at the GLE contains abundant aquaporin-4, the main water channel subtype in brain [6,7].
Corresponding author at: Center for Integrated Human Brain Science, Brain Research Institute, University of Niigata, 1 Asahimachi, Niigata 951-8585, Japan. E-mail address: [email protected] (T. Nakada). http://coe.bri.niigata-u.ac.jp.
http://dx.doi.org/10.1016/j.mri.2017.08.012 Received 2 August 2017; Accepted 31 August 2017 0730-725X/ © 2017 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
Fig. 1. Astrocyte and polarized localization of AQP-4. A: Relationship to Virchow-Robin space: Expression of AQP-4 in the brain is highly polarized to endfeet of astrocytes at two specific locations, the glia limitans externa (GLE) at the cortical surface and pericapillary Virchow-Robin space (VRS). VRS constitutes fluid-filled canals surrounding perforating arteries, capillaries and veins in the parenchyma of the brain. While the pia mater ends near the brain surface, VRS continues into the brain parenchyma with a perforating artery. B: Astrocyte and endfeet: astrocyte endfeet attach to many structures, but AQP-4 is found only at the glia limitans externa (GLE) and the Virchow-Robin space (VRS). Since AQP-4 at the capillary VRS is responsible for water efflux from astrocytes into VRS, it is highly plausible that AQP-4 at the GLE is responsible for water influx into the astrocyte from pericortical interstitial fluid space, thereby maintaining astrocyte intracellular water equilibrium.
effective TE = 141 ms; receiver bandwidth (RBW) = 15.6 kHz; echo train length (ETL) = 10; acceleration (ASSET) factor = 2; and number of excitations (NEX) = 2. An adiabatic pulse was utilized for the inversion preparation. The scan was repeated with the following series of TIs: 1000, 1400, 1900, and 2300 ms. Hereafter, the i-th TI value is denoted as TIi. Cardiac-gated FSE imaging was also performed for the same voxel locations with FSE-IR scanning to estimate T2 and apparent density of water protons (ρapp). Other specific scan parameters were as follows: TR = 5 × R–R interval; RBW = 31.25 kHz; ETL = 16; ASSET factor = 2; and NEX = 4. TE was varied as 12, 36, 72, 108, and 144 ms in a series of scans.
AQP-4 localization is highly polarized and expressed only on astrocyte endfeet at the GLE and pericapillary Virchow-Robin space (Fig. 1). Considering the anatomical location of the GLE and known AQP-4 functionality, it appears that FHR represents the GLE. The findings in this study on the MR characteristics of FHR strongly support the GLE to be the anatomical correlate of FHR. The study also supports the GLE as a structure possessing dielectric properties as previously suggested [8]. 2. Materials and methods 2.1. Subjects
2.3. Post processing Eight healthy male adults, aged 19–33 years participated in the study. The study was conducted under the tenets of the Helsinki Declaration and in accordance with the human research guidelines of the Institutional Review Board of the University of Niigata. All subjects voluntarily provided informed consent before participation in the study.
FSE-IR and cardiac-gated FSE images were subjected to spatial smoothing, affine registration, region-of-interest (ROI) selection, signal model fitting, and statistical analysis. Post-processing was performed on MATLAB version R2015b (The MathWorks, Natick, MA, USA). Spatial smoothing was done by multiplying a two-dimensional Gaussian kernel with the Fourier transform of the original slice image. Variance (σ) of the filter kernel was set to 40% of the full width of the Fourier (frequency) domain. Affine registration was accomplished utilizing the MATLAB function imregister. This function was configured for ‘multimodal’ mode with the following settings: InitialRadius = 0.009, Epsilon = 1.5e-5, GrowthFactor = 1.01, MaximumIterations = 500. A well-registered area of 128 × 128 pixels was then manually selected as the ROI for subsequent analyses, with zero-valued pixels rendered outside the region of the subarachnoid CSF. The ROI selected in this study localized in the lateral frontal cortex. The process is summarized
2.2. Data acquisition Brain MRI was performed on a 7T system (MR950; GE Healthcare, Milwaukee, WI, USA) employing a 32-channel head coil (Nova Medical, Wilmington, MA, USA). Data for T1 mapping were obtained by performing FSE-IR imaging with the following scan parameters: field of view (FOV) = 18 cm; matrix size = 512 (frequency encoding) × 256 (phase encoding), zero-filled to 512 × 512 in the reconstruction process; slice thickness (ST) = 5 mm; repetition time (TR) = 8000 ms; 141
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
Fig. 2. Flow of data analysis. The rounded rectangles represent data acquisition and processing, and the circles denote results to be discussed.
provided for easier spatial correlation of FHR in the depicted images. K-means segmentation of the obtained parameter sets quantitatively separated the image pixels into five clusters (Fig. 3, bottom row). Pictorial assessment of the distribution of each cluster, elucidated that the parameter characteristics of cluster 1, 4, and 5 reflected CSF, gray matter (GM), and white matter (WM), respectively. Similarly, the parameter set for cluster 3 most likely reflected FHR. Calculated image intensities for each cluster using parameters determined by the study were plotted against TI values (Fig. 4). The bold vertical line represented the intensities seen on actual T2-FLAIR images (TI = 1900 ms). This reproduced the actual intensity profile of T2FLAIR images for CSF, GM, WM, as well as FHR, confirming that intrinsic contrast parameters of cluster 3 reflected FHR. The results of multiple subject analysis, performed for statistical confirmation of the findings, are summarized in Fig. 5.
in Fig. 2. A T1 map was obtained by fitting the following equation to the TIdependent signal profile of each and every FSE-IR ROI pixel:
( )
( )
⎧ SFSE − IR = k ⎡1 − 2 exp − TI + exp − TR ⎤ ⎪ T1 T1 ⎦ ⎣ ⎨ TE ⎪ k = ρapp exp − T2 ⎩
( )
(1)
where k was treated as a proportionality factor. Fitting was accomplished using the MATLAB function fminsearch, which employs the Nelder-Mead simplex optimization algorithm. To account for possible negative signs in the IR signal series, fitting was repeated with five sets of signal weights {w1, w2, w3, w4, w5} = {(1 1 1 1), (− 1 1 1 1), (− 1 − 1 1 1), (− 1 − 1 − 1 1), (−1 − 1 −1 − 1)}. More concretely, the ith trial of the model fitting was done for the element-wise product of wi and s, where s = {s(TI1), s(TI2), s(TI3), s(TI4)} represents the observed magnitude signals, and the best fit among the trials was chosen as that giving the T1 estimate. Maps of ρapp and T2 were obtained by fitting the following equation to the TE-dependent signal profile of each and every cardiac-gated FSE ROI pixel:
TR ⎞ ⎤ TE ⎞ SFSE = ρapp ⎡1 − exp ⎛− exp ⎛− ⎢ ⎥ T 1 ⎠⎦ ⎝ ⎝ T2 ⎠ ⎣ ⎜
⎟
⎜
3.2. Independent component analysis ICA analysis effectively separated three independent components, referred to here as IRC 1, 2 and 3, the spatial distributions of which were pictorially analyzed (IRC map). A representative IRC map is shown in Fig. 6. Analysis of the specific distribution of each of the components realized by pictorial assessment, clearly indicated that IRC 1 (blue), IRC 2 (purple), and IRC 3 (lime) represented CSF, FHR, and GM, respectively. ICA analysis of all participants showed virtually identical results for all three independent components.
⎟
(2)
In this fitting, the parameter T1 was substituted by the value of the T1 map calculated in advance from the FSE-IR data. The estimated sets of the three intrinsic parameters (ρapp, T1, T2) were subjected to k-means clustering [9,10] to examine whether a group of pixels corresponding to FHR could be segmented. The MATLAB function kmeans was adopted for partitioning ROI pixels into a given number of groups based on the similarity between the parameter values (or geometrically, based on the closeness between 3-dimensional locations defined by the parameter sets). For multiple subject analysis, the one-tailed p-value was calculated for the difference in the intracluster means (n = 8) using the MATLAB function t-test. Independent component analysis (ICA) of FSE-IR ROI images was further performed to confirm whether T1 behavior can define FHR. The MATLAB script of FastICA algorithm [11,12] was used for this analysis. The input data were arranged in a similar way to functional MRI adopting spatial ICA [13,14]. Each of the resultant components, referred to here as inversion recovery components (IRCs), was represented by its image.
4. Discussion 4.1. Intrinsic contrast parameters of FHR Our study unambiguously showed that the intrinsic contrast parameters represented by cluster 3 reflected FHR. While the parameter values of cluster 1, 4, and 5 were all highly compatible with the known values of their corresponding structures, CSF, GM, and WM, respectively, those of cluster 3 showed distinctive characteristics, namely relatively low proton density, long T2 and disproportionally short T1. In their original article, van Veluw, et al., concluded that the major contributing factor to FHR high intensity on T2-FLAIR was long T2 [5]. Our data indicated that their conclusion was partially correct. The T2 value of cluster 3 (FHR) was indeed significantly longer (ca. 2.4 times) than that of cluster 4 (GM). Nevertheless, as is clearly appreciable in Fig. 4, T1 characteristics also play a significant role in producing the observed FHR high intensity on T2-FLAIR. Should cluster 3 retain a longer T1 similar to CSF, cluster 3 would have an intensity similar to or even lower than that of GM on T2-FLAIR. Should it acquire a shorter T1, similar to the cluster 4 or 5, cluster 3 would have an unrealistically high intensity on T2-FLAIR. The observed characteristics lead to the inescapable conclusion that the phenomenon termed FHR is conspicuous because of its unique contrast parameter characteristics, i.e., a long T2 with a disproportionally short T1. ICA is a higher order statistics based on information theory. Spatial ICA employed in this study was adopted from the method used for
3. Results 3.1. Cluster analysis The results of cluster analysis are summarized in Fig. 3. Parameter mapping of the region of interest (ROI) is displayed in the upper row. The sub-region utilized for further analysis is identified by a white square and their magnified images are presented in the middle row. The corresponding FLAIR image showing FHR is presented on the right side end of the middle row. In addition to matrix line, open arrows are 142
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
Fig. 3. Representative maps of the intrinsic contrast parameters. Top: ROI maps of apparent proton density, ρapp (left), T1 (middle), and T2 (right) estimated by pixel-by-pixel fitting of the signal models (Eqs. (1) and (2)) to the FSE-IR and FSE image series. Middle: enlarged view of a sub-region indicated by the white square on the ROI maps in the top row. FSE-IR image of TI = 1900 ms shown on the right as a typical T2-FLAIR image of the same sub-region. Bottom: five clusters of the intrinsic parameters segmented by the k-means algorithm and a list of the mean values of the intrinsic parameters within each cluster. The clusters are numbered in descending order of the mean T1. The black region of the cluster map corresponds to zero-valued image pixels defined by manual brain masking. The values of ρapp are normalized relative to cluster 1.
Fig. 4. Projected signal intensities. The curved lines show the absolute values of the FSE-IR signal model (Eq. (1)) calculated as functions of TI with substitution of the mean intrinsic parameters of the five clusters shown in Fig. 3. The bold vertical bar is positioned at TI = 1900 ms and illustrates relative signal intensities on a typical T2-FLAIR image on a 7 T system. Cluster 3 which reflects FHR is represented by the solid line.
Fig. 5. Multiple subject analysis. Multiple subject analysis confirmed the parameter characteristics discussed in Figs. 3 and 4. Each open bar represents the intra-cluster mean (n = 8). The error bar indicates standard deviations. The one tailed p-value indicates statistical significance of the difference in the mean the adjacent clusters.
correlated to FHR consisted of free water that was present in low concentrations. As discussed earlier, the GLE is the likely normal anatomical structure that gives rise to the phenomenon of FHR. The unique contrast parameters of FHR is explained when the anatomical and functional specifics of the GLE in relation to interstitial water dynamics and AQP-4 are considered. Localization of AQP-4, the main water channel in brain which is abundantly expressed at the endfeet of astrocytes, is highly polarized, existing on endfeet of astrocytes at two specific locations, namely, the pericapillary Virchow-Robin space (VRS) and GLE (Fig. 8) [6,7]. AQP-4 is responsible for water dynamics of the pericapillary VRS, believed essential for maintaining physiological interstitial flow, the lymphatic equivalent of brain known as glymphatic flow. AQP-4 realizes this function by providing water influx into the pericapillary VRS (extracellular space) from its intracellular space fluid [18]. In order to maintain intracellular water equilibrium, an equivalent amount of water needs to enter astrocytes. The likely source for water replenishment is the second site of astrocyte AQP-4 at the GLE. Here, AQP-4
functional MRI, and segregation of pixels into the same component was solely based on the behavior of T1 characteristic entirely independent from their spatial information [13,14]. As can be seen in Fig. 6, ICA effectively identified three independent components. Their distribution clearly showed that IRC 1, 2, and 3 corresponded to CSF, FHR, and GM, respectively. The ICA study results provided not only further strong support for the accuracy of the K-means segmentation and cluster assignment discussed in Fig. 3, but also clearly confirmed that T1 behavior was sufficiently distinct for identifying FHR. 4.2. Anatomical correlate of FHR FHR T2 was considerably higher than that of brain parenchyma (GM and WM), indicating that the structure which correlated to FHR had a relatively high content of unbound, free water similar to CSF. In contrast, the apparent proton density, ρapp, of FHR was significantly lower than that of CSF, at a level comparable to brain parenchyma. These unique properties strongly suggested that the structure which 143
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
similar to a contrast agent, namely interaction with a paramagnetic substrate. Although deoxyhemoglobin can display paramagnetic interaction, its strong susceptibility effects will significantly shorten apparent T2, which will be more accentuated on a higher magnetic field [4,22]. Accordingly, a structure containing deoxyhemoglobin will have significantly low intensity on a 7T system regardless of its T1 characteristics. Free oxygen molecules with unpaired electrons are another potential candidate for paramagnetic interaction. Nevertheless, it is virtually impossible for free oxygen molecules to exist in the GLE of live patients. The only viable paramagnetic candidate under physiological conditions is electrons. The brain is an active electric device where there is a constant movement of free electrons. Electron microscopy identified the GLE as an electron rich layer, indicating that this structure possesses molecular properties allowing for easy capture of free electrons [8]. Although much remains to be clarified, AQP-4 facilitated low water concentration of extracellular space at the GLE may provide the necessary electrostatic conditions to permit this electron rich status as previously suggested in relationship to Linus Pauling's aqueous phase theory for the molecular mechanism of general anesthesia [8].
Fig. 6. IRC map. The distribution maps of three IRCs decomposed from FSE-IR data based on the spatial ICA. It is clear by pictorial assessment that IRC 1, 2, and 3 correspond to CSF, FHR, and GM, respectively.
moves water from the interstitium into the intracellular space of astrocytes (Fig. 1). The functionality of AQP-4 at the GLE likely accounts for the observed relatively lower water concentration in the interstitial space of the GLE [38]. MRI evidence corroborates that the GLE is the anatomical correlate of FHR. However, the behavior of ρapp and T2 strongly suggest that the unique contrast parameters of FHR primarily reflect interstitial (extracellular) fluid rather than the intracellular fluid of astrocyte endfeet. Combining AQP-4 functionality data and these MRI findings clarifies that the GLE represents FHR macroscopically while, microscopically, FHR is represented by the extracellular (interstitials) space immediately adjacent to the GLE.
4.4. FHR analysis as index of AQP-4 functionality Fluid-filled canals surrounding perforating arteries, capillaries and veins in brain parenchyma were recognized early in modern medicine and are referred to as the Virchow-Robin space (VRS), based on the first two scientists who described the structures in detail [23,24]. It was soon identified that the interstitial fluid within the VRS has its own circulation, referred to as “interstitial flow”, which plays a role similar to systemic lymphatics for the brain which lacks a conventional lymphatic system [25–28]. The concept has recently received further attention with respect to β-amyloid clearance [21,27–30]. Accordingly, brain lymphatics, on which β-amyloid clearance is dependent, is now referred to as glymphatics, denoting glial lymphatics [31]. Water dynamics of the pericapillary VRS and, hence, glymphatic flow is dependent on functionality of AQP-4. Senile plaque formation is thought to be related to dysfunction of AQP-4 and in turn, glymphatic flow [18,21]. Direct involvement of AQP-4 in neural activities has also been implicated following the observation that deletion of α-syntrophin, a protein associated with dystrophin, in mice is associated with prolonged potassium (K+) clearance [32]. It has been shown that inwardly rectifying potassium channels, Kir4.1, responsible for the rapid removal of K+ from extracellular (interstitial) fluid essential for maintaining neural excitability, co-localizes with AQP-4 on the astrocyte endfeet facing capillary endothelium at the VRS [33]. The HCO3- independent proton pump, vacuolar ATPase (V-ATPase), expressed abundantly in astrocytes, is also considered to be a part of the dystrophin-associated glycoprotein (DCG) complex [34]. As suggested by Hibino and Kurachi [35], this co-localization may be the result of a clustering of proteins in specialized lipid raft domains which bring together functionally coupled molecules to a common site on the membrane. As in the case of neurovascular coupling, functional coupling of AQP-4, Kir4.1, and VATPase plays an essential role in maintaining brain electrical function [36]. AQP-4 positron emission tomography (PET) [37] has inherently limited spatial resolution. This limitation precludes the possibility of detailed analysis of AQP-4 functionality and detection of potential subtle abnormalities of AQP-4 under clinical settings using PET. On the other hand, detailed analysis of intrinsic imaging parameters of FHR may be developed into an index of AQP-4 functionality. Microscopic imaging using a 7T system has become clinical reality. Further investigation of FHR in various neurological pathological conditions, especially Alzheimer’s disease, (AD) is warranted.
4.3. T1 behavior of the GLE MRI contrast parameters of the GLE can be readily explained by its unique extracellular space fluid condition, namely, free water in low concentration. A drawback of this conceptualization is the observed T1 behavior which cannot simply be realized based on free water of the interstitial space present in relatively low concentration. It is known that 30 mg/dl of albumin, corresponding to a normal CSF protein level, reduces the T1 and T2 of the solution by 5 and 20% (compared to pure saline), respectively [19]. The T1 and T2 shortening ratios of CSF under the highest protein concentration, 4000 mg/dl in plasma, are approximately 20 and 65%, respectively [20]. In other words, the effect of protein is always more pronounced for T2 rather than T1. In this context, the observed T1 shortening of ca. 33% is highly disproportionate considering the actual observed T2 shortening of only ca. 33% (T2 of the GLE is ca. 66% that of CSF). One must conclude that there is another factor at play responsible for the degree of observed T1 shortening. Biologically compatible agents well recognized to shorten T1 disproportionately more than T2 are paramagnetic contrast agents [20]. Assuming, without a contrast agent, T1 = 3 s and T2 = 0.3 s and specific relaxivities of rC,1 = rC,2 = 0.2, relaxation times can be expressed as follows [20]:
T1 = (r1 + rC,1)−1 = (1 3 + 0.2)−1 = 1.875 s
T2 = (r2 + rC,2)−1 = (1 0.3 + 0.2)−1 = 0.283 s. The reduction ratios are then calculated to be:
%ΔT1 = 100 × (3 − 1.875) 3 = 37.5%
%ΔT2 = 100 × (0.3 − 0.283) 0.3 = 5.7% Therefore, it appears that T1 of the GLE demonstrates T1 shortening 144
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
[14] Nakada T, Suzuki K, Fujii Y, Matsuzawa H, Kwee IL. Independent component-cross correlation-sequential epoch (ICS) analysis of high-field fMRI time series: direct visualization of dual representation of the primary motor cortex in human. Neurosci. Res. 2000;37:237–44. [18] Haj-Yasein NN, Jensen V, Østby I, Omholt SW, Voipio J, Kaila K, et al. Aquaporin-4 regulates extracellular space volume dynamics during high-frequency synaptic stimulation: a gene deletion study in mouse hippocampus. Glia 2012;60:867–74. [19] Melhem ER, Jara H, Eustace S. Fluid-attenuated inversion recovery MR imaging: identification of protein concentration thresholds for CSF hyperintensity. Am. J. Roentgenol. 1997;169:859–62. [20] Castro ME, Boisvert DP, Treiber EO, Lunt JA, Allen PS. Effect of CSF albumin concentration on NMR relaxation parameters. In: Inaba Y, Klatzo I, Spatz M, editors. Brain Edema. Berlin Heidelberg: Springer-Verlag; 1985. p. 594–600. [21] Nakada T. Virchow-Robin space and aquaporin-4: new insights on an old friend. Croat. Med. J. 2014;55:328–36. [22] Rohrer M, Bauer H, Mintorovitch J, Requardt M, Weinmann HJ. Comparison of magnetic properties of MRI contrast media solutions at different magnetic field strengths. Investig. Radiol. 2005;40:715–24. [23] Virchow R. Ueber die Erweiterung kleinerer Gefaesse. Arch. Pathol. Anat. Physiol. Klin. Med. 1851;3:427–62. [24] Robin C. Recherches sur quelques particularites de la structure des capillaires de l’encephale. J. Physiol. Homme Animaux 1859;2:537–48. [25] Esiri MM, Gay D. Immunological and neuropathological significance of the Virchow-Robin space. J. Neurol. Sci. 1990;100:3–8. [26] Weller RO. Pathology of cerebrospinal fluid and interstitial fluid of the CNS: significance for Alzheimer's disease, prion disorders and multiple sclerosis. J. Neuropathol. Exp. Neurol. 1998;57:885–94. [27] Johnston M, Papaiconomou C. Cerebrospinal fluid transport: a lymphatic perspective. News Physiol. Sci. 2002;17:227–30. [28] Abbott NJ. Evidence for bulk flow of brain interstitial fluid: significance for physiology and pathology. Neurochem. Int. 2004;45:545–52. [29] Igarashi H, Suzuki Y, Kwee IL, Nakada T. Water influx into cerebrospinal fluid is significantly reduced in senile plaque bearing transgenic mice, supporting ß-amyloid clearance hypothesis of Alzheimer disease. Neurol. Res. 2014;36:1094–8. [30] Suzuki Y, Nakamura Y, Yamada K, Igarashi H, Kasuga K, Yokoyama Y, et al. Reduced CSF water influx in Alzheimer's disease supporting the ß-amyloid clearance hypothesis. PLoS One 2015;10:e0123708. [31] Iliff JJ, Wang M, Liao Y, Plogg BA, Peng W, Gundersen GA, et al. A paravascular pathway facilitates CSF flow through the brain parenchyma and the clearance of interstitial solutes, including amyloid ß. Sci. Transl. Med. 2012;4. (147ra111). [32] Amiry-Moghaddam M, Williamson A, Palomba M, Eid T, de Lanerolle NC, Nagelhus EA, et al. Delayed K + clearance associated with aquaporin-4 mislocalization: phenotypic defects in brains of α-syntrophin-null mice. Proc. Natl. Acad. Sci. U. S. A. 2003;100:13615–20. [33] Olsen ML, Sontheimer H. Functional implications for Kir4.1 channels in glial biology: from K + buffering to cell differentiation. Neurochem. 2008;107:589–601. [34] Chen J, Skinner MA, Shi W, Yu Q-C, Wildeman AG, Chan YM. The 16 kDa subunit of vacuolar H +-ATPase is a novel sarcoglycan-interacting protein. Biochim. Biophys. Acta 1772;2007:570–9. [35] Hibino H, Kurachi Y. Distinct detergent-resistant membrane microdomains (lipid rafts) respectively harvest K(+) and water transport systems in brain astroglia. Eur. J. Neurosci. 2007;26:2539–55. [36] Nakada T. The molecular mechanisms of neural flow coupling: a new concept. J Neuroimaging 2015;25:861–5. [37] Suzuki Y, Nakamura N, Yamada K, Huber VJ, Mika Tsujita M, Nakada T. Aquaporin4 positron emission tomography imaging of the human brain: first report. J. Neuroimaging 2013;23:219–23. [38] Nakada T, Kwee IL, Igarashi H, Suzuki Y. Aquaporin-4 functionality and VirchowRobin space water dynamics: physiological model for neurovascular coupling and glymphatic flow. Int. J. Mol. Sci. 2017;18:1798.
5. Conclusion The results of this study strongly support the anatomical substrate of FHR identified on a 7T system to be the GLE and its immediately adjacent interstitial space. The values of the GLE’s intrinsic contrast parameters, ρapp, T1, T2, show unique characteristics compatible with the structural and functional specifics of the GLE. The GLE is composed of snuggled endfeet of astrocytes with abundant AQP-4, the main water channel in brain. AQP-4 plays an essential role in the water dynamics of the brain inside the blood brain barrier. Since the unique combination of contrast parameters of FHR is strongly dependent on functionality of AQP-4, microscopic imaging of the GLE and analysis of its contrast parameters can be developed into an index of AQP-4 functionality at the GLE, and its possible role in neurological diseases. Acknowledgements This study is supported by grants from the Ministry of Education, Culture, Sports, Science, and Technology (Japan) and University of Niigata. The study was in part presented at the Society of Neuroscience 2015, 578.05/C23. References [1] Adams JG, Melhem ER. Clinical usefulness of T2-weighted fluid-attenuated inversion recovery MR imaging of the CNS. Am. J. Roentgenol. 1999;172:529–36. [2] Stuckey SL, Goh TD, Heffernan T, Rowan D. Hyperintensity in the subarachnoid space on FLAIR MRI. Am. J. Roentgenol. 2007;189:913–21. [3] Nakada T, Matsuzawa H, Kwee IL. High resolution imaging with high and ultrahigh-field MRI systems. NeuroReport 2008;19:7–13. [4] Nakada T, Matsuzawa H, Igarashi H, Fujii Y, Kwee IL. In vivo visualization of senile plaque like pathology in Alzheimer's Disease patients by MR microscopy on a 7T system. J. Neuroimaging 2008;18:125–9. [5] van Veluw SJ, Fracasso A, Visser F, Spliet WG, Luijten PR, Biessels GJ, et al. FLAIR images at 7 Tesla MRI highlight the ependyma and the outer layers of the cerebral cortex. NeuroImage 2015;104:100–9. [6] Rash JE, Yasumura T, Hudson CS, Agre P, Nielsen S. Direct immunogold labeling of aquaporin-4 in square arrays of astrocyte and ependymocyte plasma membranes in rat brain and spinal cord. Proc. Natl. Acad. Sci. U. S. A. 1998;95:11981–6. [7] Neely JD, Christensen BM, Nielsen S, Agre P. Heterotetrameric composition of aquaporin-4 water channels. Biochemistry 1999;38:11156–63. [8] Nakada T. Neuroscience of water molecules: a salute to Professor Linus Carl Pauling. Cytotechnology 2009;59:145–52. [9] MacQueen JB. Some methods for classification and analysis of multivariate observations. Proc Fifth Berkeley Symp on Math Statist and Prob. 1. 1967. p. 281297. [10] Vijayalakshmi P, Selvamani K, Geetha M. Segmentation of brain MRI using k-means clustering algorithm. Int. J. Engl. Trends Technol. 2011;3:113–5. [11] Hyvärinen A. Fast and robust fixed-point algorithms for independent component analysis. IEEE Trans. Neural Netw. 1999;10:626–34. [12] http://research.ics.aalto.fi/ica/fastica/. [13] McKeown MJ, Jung TP, Makeig S, Brown G, Kindermann SS, Lee TW, et al. Spatially independent activity patterns in functional MRI data during the stroop colornaming task. Proc. Natl. Acad. Sci. U. S. A. 1998;95:803–10.
145
Contents lists available at ScienceDirect
Magnetic Resonance Imaging journal homepage: www.elsevier.com/locate/mri
MRI characteristics of the glia limitans externa: A 7T study a
a
a
a
MARK a
Kiyotaka Suzuki , Kenichi Yamada , Kazunori Nakada , Yuji Suzuki , Masaki Watanabe , Ingrid L. Kweea,b, Tsutomu Nakadaa,b,⁎,1 a b
Center for Integrated Human Brain Science, Brain Research Institute, University of Niigata, Niigata, Japan Department of Neurology, University of California, Davis, USA
A R T I C L E I N F O
A B S T R A C T
Keywords: FLAIR ICA k-means clustering AQP-4
Purpose: To perform a systematic analysis of the intrinsic contrast parameters of the FLAIR hyperintense rim (FHR), a thin layer of high intensity covering the entire surface of the cerebral cortex detected on fluid-attenuated inversion recovery (FLAIR) sequence T2 weighted imaging performed on a 7T system, in an attempt to identify its anatomical correlate. Methods: Fast spin echo inversion recovery (FSE-IR) and cardiac-gated fast spin echo (FSE) images were obtained with defined parameters in eight normal volunteers on a 7 T MRI system to determine T2 and proton density, T1 characteristics. K-means clustering analysis of parameter sets was performed using MATLAB version R2015b for the purpose of identifying the cluster reflecting FHR. The results were subsequently confirmed by independent component analysis (ICA) based on T1 behavior on FSE-IR using a MATLAB script of FastICA algorithm. Results: The structure giving rise to FHR was found to have a unique combination of intrinsic contrast parameters of low proton density, long T2, and disproportionally short T1. The findings are in strong agreement with the functional and structural specifics of the glia limitans externa (GLE), a structure composed of snuggled endfeet of astrocytes containing abundant aquaporin-4 (AQP-4), the main water channel of the brain. Conclusion: Intrinsic contrast parameters of FHR reflect structural and functional specifics of the GLE, and their values are highly dependent on the physiologic functionality of AQP-4. Microscopic imaging on a 7T system and analysis of GLE contrast parameters can be developed into a method for evaluating AQP-4 functionality.
1. Introduction Fluid-attenuated inversion recovery (FLAIR) sequence with T2 (transverse or spin-spin relaxation time) weighting, T2-FLAIR, is a widely used magnetic resonance imaging (MRI) sequence for detecting structural brain abnormalities. The sequence is designed to suppress background high intensity of cerebrospinal fluid (CSF) on T2 weighted images by optimizing inversion time (TI) to the zero-crossing point of the longitudinal magnetization of CSF. T2-FLAIR imaging has made possible the detection of subtle abnormalities of structures adjacent to the cerebral cortex [1]. Detection of high intensity lesions on the surface of the brain and within the CSF space by T2-FLAIR on conventional MRI systems have been thought to reflect pathological conditions or artifacts [2]. MRI technology has recently entered the high and ultra-high field era with higher resolution images on 3 to 7 Tesla systems, and has led to new findings of the human cortex [3–5]. One unexpected finding is the
⁎
1
high intensities in a thin layer of normal cortex identified with T2FLAIR imaging on a 7T system. The unusual phenomenon has been referred to as “FLAIR hyperintense rim (FHR)” [5]. It is believed that FHR reflects a normal anatomical structure on the surface of the cerebral cortex. This structure possesses significantly long T2, but it is nevertheless distinct from CSF. Thus far, there have been no published studies in the English literature on the systematic evaluation of the structure corresponding to FHR. The surface of the cerebral cortex adjacent to the CSF filled subarachnoid space is enveloped by the pia matter, a thin fibrous tissue that is impermeable to fluid. This simple structure has no known properties that would lead to the unique MRI characteristic responsible for FHR. Immediately beneath the pia matter, however, is a ribbon like structure, known as the glia limitans externa (GLE), which has several unique anatomical characteristics. First, the GLE is composed of snuggled endfeet of astrocytes. Second, astrocyte endfeet at the GLE contains abundant aquaporin-4, the main water channel subtype in brain [6,7].
Corresponding author at: Center for Integrated Human Brain Science, Brain Research Institute, University of Niigata, 1 Asahimachi, Niigata 951-8585, Japan. E-mail address: [email protected] (T. Nakada). http://coe.bri.niigata-u.ac.jp.
http://dx.doi.org/10.1016/j.mri.2017.08.012 Received 2 August 2017; Accepted 31 August 2017 0730-725X/ © 2017 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
Fig. 1. Astrocyte and polarized localization of AQP-4. A: Relationship to Virchow-Robin space: Expression of AQP-4 in the brain is highly polarized to endfeet of astrocytes at two specific locations, the glia limitans externa (GLE) at the cortical surface and pericapillary Virchow-Robin space (VRS). VRS constitutes fluid-filled canals surrounding perforating arteries, capillaries and veins in the parenchyma of the brain. While the pia mater ends near the brain surface, VRS continues into the brain parenchyma with a perforating artery. B: Astrocyte and endfeet: astrocyte endfeet attach to many structures, but AQP-4 is found only at the glia limitans externa (GLE) and the Virchow-Robin space (VRS). Since AQP-4 at the capillary VRS is responsible for water efflux from astrocytes into VRS, it is highly plausible that AQP-4 at the GLE is responsible for water influx into the astrocyte from pericortical interstitial fluid space, thereby maintaining astrocyte intracellular water equilibrium.
effective TE = 141 ms; receiver bandwidth (RBW) = 15.6 kHz; echo train length (ETL) = 10; acceleration (ASSET) factor = 2; and number of excitations (NEX) = 2. An adiabatic pulse was utilized for the inversion preparation. The scan was repeated with the following series of TIs: 1000, 1400, 1900, and 2300 ms. Hereafter, the i-th TI value is denoted as TIi. Cardiac-gated FSE imaging was also performed for the same voxel locations with FSE-IR scanning to estimate T2 and apparent density of water protons (ρapp). Other specific scan parameters were as follows: TR = 5 × R–R interval; RBW = 31.25 kHz; ETL = 16; ASSET factor = 2; and NEX = 4. TE was varied as 12, 36, 72, 108, and 144 ms in a series of scans.
AQP-4 localization is highly polarized and expressed only on astrocyte endfeet at the GLE and pericapillary Virchow-Robin space (Fig. 1). Considering the anatomical location of the GLE and known AQP-4 functionality, it appears that FHR represents the GLE. The findings in this study on the MR characteristics of FHR strongly support the GLE to be the anatomical correlate of FHR. The study also supports the GLE as a structure possessing dielectric properties as previously suggested [8]. 2. Materials and methods 2.1. Subjects
2.3. Post processing Eight healthy male adults, aged 19–33 years participated in the study. The study was conducted under the tenets of the Helsinki Declaration and in accordance with the human research guidelines of the Institutional Review Board of the University of Niigata. All subjects voluntarily provided informed consent before participation in the study.
FSE-IR and cardiac-gated FSE images were subjected to spatial smoothing, affine registration, region-of-interest (ROI) selection, signal model fitting, and statistical analysis. Post-processing was performed on MATLAB version R2015b (The MathWorks, Natick, MA, USA). Spatial smoothing was done by multiplying a two-dimensional Gaussian kernel with the Fourier transform of the original slice image. Variance (σ) of the filter kernel was set to 40% of the full width of the Fourier (frequency) domain. Affine registration was accomplished utilizing the MATLAB function imregister. This function was configured for ‘multimodal’ mode with the following settings: InitialRadius = 0.009, Epsilon = 1.5e-5, GrowthFactor = 1.01, MaximumIterations = 500. A well-registered area of 128 × 128 pixels was then manually selected as the ROI for subsequent analyses, with zero-valued pixels rendered outside the region of the subarachnoid CSF. The ROI selected in this study localized in the lateral frontal cortex. The process is summarized
2.2. Data acquisition Brain MRI was performed on a 7T system (MR950; GE Healthcare, Milwaukee, WI, USA) employing a 32-channel head coil (Nova Medical, Wilmington, MA, USA). Data for T1 mapping were obtained by performing FSE-IR imaging with the following scan parameters: field of view (FOV) = 18 cm; matrix size = 512 (frequency encoding) × 256 (phase encoding), zero-filled to 512 × 512 in the reconstruction process; slice thickness (ST) = 5 mm; repetition time (TR) = 8000 ms; 141
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
Fig. 2. Flow of data analysis. The rounded rectangles represent data acquisition and processing, and the circles denote results to be discussed.
provided for easier spatial correlation of FHR in the depicted images. K-means segmentation of the obtained parameter sets quantitatively separated the image pixels into five clusters (Fig. 3, bottom row). Pictorial assessment of the distribution of each cluster, elucidated that the parameter characteristics of cluster 1, 4, and 5 reflected CSF, gray matter (GM), and white matter (WM), respectively. Similarly, the parameter set for cluster 3 most likely reflected FHR. Calculated image intensities for each cluster using parameters determined by the study were plotted against TI values (Fig. 4). The bold vertical line represented the intensities seen on actual T2-FLAIR images (TI = 1900 ms). This reproduced the actual intensity profile of T2FLAIR images for CSF, GM, WM, as well as FHR, confirming that intrinsic contrast parameters of cluster 3 reflected FHR. The results of multiple subject analysis, performed for statistical confirmation of the findings, are summarized in Fig. 5.
in Fig. 2. A T1 map was obtained by fitting the following equation to the TIdependent signal profile of each and every FSE-IR ROI pixel:
( )
( )
⎧ SFSE − IR = k ⎡1 − 2 exp − TI + exp − TR ⎤ ⎪ T1 T1 ⎦ ⎣ ⎨ TE ⎪ k = ρapp exp − T2 ⎩
( )
(1)
where k was treated as a proportionality factor. Fitting was accomplished using the MATLAB function fminsearch, which employs the Nelder-Mead simplex optimization algorithm. To account for possible negative signs in the IR signal series, fitting was repeated with five sets of signal weights {w1, w2, w3, w4, w5} = {(1 1 1 1), (− 1 1 1 1), (− 1 − 1 1 1), (− 1 − 1 − 1 1), (−1 − 1 −1 − 1)}. More concretely, the ith trial of the model fitting was done for the element-wise product of wi and s, where s = {s(TI1), s(TI2), s(TI3), s(TI4)} represents the observed magnitude signals, and the best fit among the trials was chosen as that giving the T1 estimate. Maps of ρapp and T2 were obtained by fitting the following equation to the TE-dependent signal profile of each and every cardiac-gated FSE ROI pixel:
TR ⎞ ⎤ TE ⎞ SFSE = ρapp ⎡1 − exp ⎛− exp ⎛− ⎢ ⎥ T 1 ⎠⎦ ⎝ ⎝ T2 ⎠ ⎣ ⎜
⎟
⎜
3.2. Independent component analysis ICA analysis effectively separated three independent components, referred to here as IRC 1, 2 and 3, the spatial distributions of which were pictorially analyzed (IRC map). A representative IRC map is shown in Fig. 6. Analysis of the specific distribution of each of the components realized by pictorial assessment, clearly indicated that IRC 1 (blue), IRC 2 (purple), and IRC 3 (lime) represented CSF, FHR, and GM, respectively. ICA analysis of all participants showed virtually identical results for all three independent components.
⎟
(2)
In this fitting, the parameter T1 was substituted by the value of the T1 map calculated in advance from the FSE-IR data. The estimated sets of the three intrinsic parameters (ρapp, T1, T2) were subjected to k-means clustering [9,10] to examine whether a group of pixels corresponding to FHR could be segmented. The MATLAB function kmeans was adopted for partitioning ROI pixels into a given number of groups based on the similarity between the parameter values (or geometrically, based on the closeness between 3-dimensional locations defined by the parameter sets). For multiple subject analysis, the one-tailed p-value was calculated for the difference in the intracluster means (n = 8) using the MATLAB function t-test. Independent component analysis (ICA) of FSE-IR ROI images was further performed to confirm whether T1 behavior can define FHR. The MATLAB script of FastICA algorithm [11,12] was used for this analysis. The input data were arranged in a similar way to functional MRI adopting spatial ICA [13,14]. Each of the resultant components, referred to here as inversion recovery components (IRCs), was represented by its image.
4. Discussion 4.1. Intrinsic contrast parameters of FHR Our study unambiguously showed that the intrinsic contrast parameters represented by cluster 3 reflected FHR. While the parameter values of cluster 1, 4, and 5 were all highly compatible with the known values of their corresponding structures, CSF, GM, and WM, respectively, those of cluster 3 showed distinctive characteristics, namely relatively low proton density, long T2 and disproportionally short T1. In their original article, van Veluw, et al., concluded that the major contributing factor to FHR high intensity on T2-FLAIR was long T2 [5]. Our data indicated that their conclusion was partially correct. The T2 value of cluster 3 (FHR) was indeed significantly longer (ca. 2.4 times) than that of cluster 4 (GM). Nevertheless, as is clearly appreciable in Fig. 4, T1 characteristics also play a significant role in producing the observed FHR high intensity on T2-FLAIR. Should cluster 3 retain a longer T1 similar to CSF, cluster 3 would have an intensity similar to or even lower than that of GM on T2-FLAIR. Should it acquire a shorter T1, similar to the cluster 4 or 5, cluster 3 would have an unrealistically high intensity on T2-FLAIR. The observed characteristics lead to the inescapable conclusion that the phenomenon termed FHR is conspicuous because of its unique contrast parameter characteristics, i.e., a long T2 with a disproportionally short T1. ICA is a higher order statistics based on information theory. Spatial ICA employed in this study was adopted from the method used for
3. Results 3.1. Cluster analysis The results of cluster analysis are summarized in Fig. 3. Parameter mapping of the region of interest (ROI) is displayed in the upper row. The sub-region utilized for further analysis is identified by a white square and their magnified images are presented in the middle row. The corresponding FLAIR image showing FHR is presented on the right side end of the middle row. In addition to matrix line, open arrows are 142
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
Fig. 3. Representative maps of the intrinsic contrast parameters. Top: ROI maps of apparent proton density, ρapp (left), T1 (middle), and T2 (right) estimated by pixel-by-pixel fitting of the signal models (Eqs. (1) and (2)) to the FSE-IR and FSE image series. Middle: enlarged view of a sub-region indicated by the white square on the ROI maps in the top row. FSE-IR image of TI = 1900 ms shown on the right as a typical T2-FLAIR image of the same sub-region. Bottom: five clusters of the intrinsic parameters segmented by the k-means algorithm and a list of the mean values of the intrinsic parameters within each cluster. The clusters are numbered in descending order of the mean T1. The black region of the cluster map corresponds to zero-valued image pixels defined by manual brain masking. The values of ρapp are normalized relative to cluster 1.
Fig. 4. Projected signal intensities. The curved lines show the absolute values of the FSE-IR signal model (Eq. (1)) calculated as functions of TI with substitution of the mean intrinsic parameters of the five clusters shown in Fig. 3. The bold vertical bar is positioned at TI = 1900 ms and illustrates relative signal intensities on a typical T2-FLAIR image on a 7 T system. Cluster 3 which reflects FHR is represented by the solid line.
Fig. 5. Multiple subject analysis. Multiple subject analysis confirmed the parameter characteristics discussed in Figs. 3 and 4. Each open bar represents the intra-cluster mean (n = 8). The error bar indicates standard deviations. The one tailed p-value indicates statistical significance of the difference in the mean the adjacent clusters.
correlated to FHR consisted of free water that was present in low concentrations. As discussed earlier, the GLE is the likely normal anatomical structure that gives rise to the phenomenon of FHR. The unique contrast parameters of FHR is explained when the anatomical and functional specifics of the GLE in relation to interstitial water dynamics and AQP-4 are considered. Localization of AQP-4, the main water channel in brain which is abundantly expressed at the endfeet of astrocytes, is highly polarized, existing on endfeet of astrocytes at two specific locations, namely, the pericapillary Virchow-Robin space (VRS) and GLE (Fig. 8) [6,7]. AQP-4 is responsible for water dynamics of the pericapillary VRS, believed essential for maintaining physiological interstitial flow, the lymphatic equivalent of brain known as glymphatic flow. AQP-4 realizes this function by providing water influx into the pericapillary VRS (extracellular space) from its intracellular space fluid [18]. In order to maintain intracellular water equilibrium, an equivalent amount of water needs to enter astrocytes. The likely source for water replenishment is the second site of astrocyte AQP-4 at the GLE. Here, AQP-4
functional MRI, and segregation of pixels into the same component was solely based on the behavior of T1 characteristic entirely independent from their spatial information [13,14]. As can be seen in Fig. 6, ICA effectively identified three independent components. Their distribution clearly showed that IRC 1, 2, and 3 corresponded to CSF, FHR, and GM, respectively. The ICA study results provided not only further strong support for the accuracy of the K-means segmentation and cluster assignment discussed in Fig. 3, but also clearly confirmed that T1 behavior was sufficiently distinct for identifying FHR. 4.2. Anatomical correlate of FHR FHR T2 was considerably higher than that of brain parenchyma (GM and WM), indicating that the structure which correlated to FHR had a relatively high content of unbound, free water similar to CSF. In contrast, the apparent proton density, ρapp, of FHR was significantly lower than that of CSF, at a level comparable to brain parenchyma. These unique properties strongly suggested that the structure which 143
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
similar to a contrast agent, namely interaction with a paramagnetic substrate. Although deoxyhemoglobin can display paramagnetic interaction, its strong susceptibility effects will significantly shorten apparent T2, which will be more accentuated on a higher magnetic field [4,22]. Accordingly, a structure containing deoxyhemoglobin will have significantly low intensity on a 7T system regardless of its T1 characteristics. Free oxygen molecules with unpaired electrons are another potential candidate for paramagnetic interaction. Nevertheless, it is virtually impossible for free oxygen molecules to exist in the GLE of live patients. The only viable paramagnetic candidate under physiological conditions is electrons. The brain is an active electric device where there is a constant movement of free electrons. Electron microscopy identified the GLE as an electron rich layer, indicating that this structure possesses molecular properties allowing for easy capture of free electrons [8]. Although much remains to be clarified, AQP-4 facilitated low water concentration of extracellular space at the GLE may provide the necessary electrostatic conditions to permit this electron rich status as previously suggested in relationship to Linus Pauling's aqueous phase theory for the molecular mechanism of general anesthesia [8].
Fig. 6. IRC map. The distribution maps of three IRCs decomposed from FSE-IR data based on the spatial ICA. It is clear by pictorial assessment that IRC 1, 2, and 3 correspond to CSF, FHR, and GM, respectively.
moves water from the interstitium into the intracellular space of astrocytes (Fig. 1). The functionality of AQP-4 at the GLE likely accounts for the observed relatively lower water concentration in the interstitial space of the GLE [38]. MRI evidence corroborates that the GLE is the anatomical correlate of FHR. However, the behavior of ρapp and T2 strongly suggest that the unique contrast parameters of FHR primarily reflect interstitial (extracellular) fluid rather than the intracellular fluid of astrocyte endfeet. Combining AQP-4 functionality data and these MRI findings clarifies that the GLE represents FHR macroscopically while, microscopically, FHR is represented by the extracellular (interstitials) space immediately adjacent to the GLE.
4.4. FHR analysis as index of AQP-4 functionality Fluid-filled canals surrounding perforating arteries, capillaries and veins in brain parenchyma were recognized early in modern medicine and are referred to as the Virchow-Robin space (VRS), based on the first two scientists who described the structures in detail [23,24]. It was soon identified that the interstitial fluid within the VRS has its own circulation, referred to as “interstitial flow”, which plays a role similar to systemic lymphatics for the brain which lacks a conventional lymphatic system [25–28]. The concept has recently received further attention with respect to β-amyloid clearance [21,27–30]. Accordingly, brain lymphatics, on which β-amyloid clearance is dependent, is now referred to as glymphatics, denoting glial lymphatics [31]. Water dynamics of the pericapillary VRS and, hence, glymphatic flow is dependent on functionality of AQP-4. Senile plaque formation is thought to be related to dysfunction of AQP-4 and in turn, glymphatic flow [18,21]. Direct involvement of AQP-4 in neural activities has also been implicated following the observation that deletion of α-syntrophin, a protein associated with dystrophin, in mice is associated with prolonged potassium (K+) clearance [32]. It has been shown that inwardly rectifying potassium channels, Kir4.1, responsible for the rapid removal of K+ from extracellular (interstitial) fluid essential for maintaining neural excitability, co-localizes with AQP-4 on the astrocyte endfeet facing capillary endothelium at the VRS [33]. The HCO3- independent proton pump, vacuolar ATPase (V-ATPase), expressed abundantly in astrocytes, is also considered to be a part of the dystrophin-associated glycoprotein (DCG) complex [34]. As suggested by Hibino and Kurachi [35], this co-localization may be the result of a clustering of proteins in specialized lipid raft domains which bring together functionally coupled molecules to a common site on the membrane. As in the case of neurovascular coupling, functional coupling of AQP-4, Kir4.1, and VATPase plays an essential role in maintaining brain electrical function [36]. AQP-4 positron emission tomography (PET) [37] has inherently limited spatial resolution. This limitation precludes the possibility of detailed analysis of AQP-4 functionality and detection of potential subtle abnormalities of AQP-4 under clinical settings using PET. On the other hand, detailed analysis of intrinsic imaging parameters of FHR may be developed into an index of AQP-4 functionality. Microscopic imaging using a 7T system has become clinical reality. Further investigation of FHR in various neurological pathological conditions, especially Alzheimer’s disease, (AD) is warranted.
4.3. T1 behavior of the GLE MRI contrast parameters of the GLE can be readily explained by its unique extracellular space fluid condition, namely, free water in low concentration. A drawback of this conceptualization is the observed T1 behavior which cannot simply be realized based on free water of the interstitial space present in relatively low concentration. It is known that 30 mg/dl of albumin, corresponding to a normal CSF protein level, reduces the T1 and T2 of the solution by 5 and 20% (compared to pure saline), respectively [19]. The T1 and T2 shortening ratios of CSF under the highest protein concentration, 4000 mg/dl in plasma, are approximately 20 and 65%, respectively [20]. In other words, the effect of protein is always more pronounced for T2 rather than T1. In this context, the observed T1 shortening of ca. 33% is highly disproportionate considering the actual observed T2 shortening of only ca. 33% (T2 of the GLE is ca. 66% that of CSF). One must conclude that there is another factor at play responsible for the degree of observed T1 shortening. Biologically compatible agents well recognized to shorten T1 disproportionately more than T2 are paramagnetic contrast agents [20]. Assuming, without a contrast agent, T1 = 3 s and T2 = 0.3 s and specific relaxivities of rC,1 = rC,2 = 0.2, relaxation times can be expressed as follows [20]:
T1 = (r1 + rC,1)−1 = (1 3 + 0.2)−1 = 1.875 s
T2 = (r2 + rC,2)−1 = (1 0.3 + 0.2)−1 = 0.283 s. The reduction ratios are then calculated to be:
%ΔT1 = 100 × (3 − 1.875) 3 = 37.5%
%ΔT2 = 100 × (0.3 − 0.283) 0.3 = 5.7% Therefore, it appears that T1 of the GLE demonstrates T1 shortening 144
Magnetic Resonance Imaging 44 (2017) 140–145
K. Suzuki et al.
[14] Nakada T, Suzuki K, Fujii Y, Matsuzawa H, Kwee IL. Independent component-cross correlation-sequential epoch (ICS) analysis of high-field fMRI time series: direct visualization of dual representation of the primary motor cortex in human. Neurosci. Res. 2000;37:237–44. [18] Haj-Yasein NN, Jensen V, Østby I, Omholt SW, Voipio J, Kaila K, et al. Aquaporin-4 regulates extracellular space volume dynamics during high-frequency synaptic stimulation: a gene deletion study in mouse hippocampus. Glia 2012;60:867–74. [19] Melhem ER, Jara H, Eustace S. Fluid-attenuated inversion recovery MR imaging: identification of protein concentration thresholds for CSF hyperintensity. Am. J. Roentgenol. 1997;169:859–62. [20] Castro ME, Boisvert DP, Treiber EO, Lunt JA, Allen PS. Effect of CSF albumin concentration on NMR relaxation parameters. In: Inaba Y, Klatzo I, Spatz M, editors. Brain Edema. Berlin Heidelberg: Springer-Verlag; 1985. p. 594–600. [21] Nakada T. Virchow-Robin space and aquaporin-4: new insights on an old friend. Croat. Med. J. 2014;55:328–36. [22] Rohrer M, Bauer H, Mintorovitch J, Requardt M, Weinmann HJ. Comparison of magnetic properties of MRI contrast media solutions at different magnetic field strengths. Investig. Radiol. 2005;40:715–24. [23] Virchow R. Ueber die Erweiterung kleinerer Gefaesse. Arch. Pathol. Anat. Physiol. Klin. Med. 1851;3:427–62. [24] Robin C. Recherches sur quelques particularites de la structure des capillaires de l’encephale. J. Physiol. Homme Animaux 1859;2:537–48. [25] Esiri MM, Gay D. Immunological and neuropathological significance of the Virchow-Robin space. J. Neurol. Sci. 1990;100:3–8. [26] Weller RO. Pathology of cerebrospinal fluid and interstitial fluid of the CNS: significance for Alzheimer's disease, prion disorders and multiple sclerosis. J. Neuropathol. Exp. Neurol. 1998;57:885–94. [27] Johnston M, Papaiconomou C. Cerebrospinal fluid transport: a lymphatic perspective. News Physiol. Sci. 2002;17:227–30. [28] Abbott NJ. Evidence for bulk flow of brain interstitial fluid: significance for physiology and pathology. Neurochem. Int. 2004;45:545–52. [29] Igarashi H, Suzuki Y, Kwee IL, Nakada T. Water influx into cerebrospinal fluid is significantly reduced in senile plaque bearing transgenic mice, supporting ß-amyloid clearance hypothesis of Alzheimer disease. Neurol. Res. 2014;36:1094–8. [30] Suzuki Y, Nakamura Y, Yamada K, Igarashi H, Kasuga K, Yokoyama Y, et al. Reduced CSF water influx in Alzheimer's disease supporting the ß-amyloid clearance hypothesis. PLoS One 2015;10:e0123708. [31] Iliff JJ, Wang M, Liao Y, Plogg BA, Peng W, Gundersen GA, et al. A paravascular pathway facilitates CSF flow through the brain parenchyma and the clearance of interstitial solutes, including amyloid ß. Sci. Transl. Med. 2012;4. (147ra111). [32] Amiry-Moghaddam M, Williamson A, Palomba M, Eid T, de Lanerolle NC, Nagelhus EA, et al. Delayed K + clearance associated with aquaporin-4 mislocalization: phenotypic defects in brains of α-syntrophin-null mice. Proc. Natl. Acad. Sci. U. S. A. 2003;100:13615–20. [33] Olsen ML, Sontheimer H. Functional implications for Kir4.1 channels in glial biology: from K + buffering to cell differentiation. Neurochem. 2008;107:589–601. [34] Chen J, Skinner MA, Shi W, Yu Q-C, Wildeman AG, Chan YM. The 16 kDa subunit of vacuolar H +-ATPase is a novel sarcoglycan-interacting protein. Biochim. Biophys. Acta 1772;2007:570–9. [35] Hibino H, Kurachi Y. Distinct detergent-resistant membrane microdomains (lipid rafts) respectively harvest K(+) and water transport systems in brain astroglia. Eur. J. Neurosci. 2007;26:2539–55. [36] Nakada T. The molecular mechanisms of neural flow coupling: a new concept. J Neuroimaging 2015;25:861–5. [37] Suzuki Y, Nakamura N, Yamada K, Huber VJ, Mika Tsujita M, Nakada T. Aquaporin4 positron emission tomography imaging of the human brain: first report. J. Neuroimaging 2013;23:219–23. [38] Nakada T, Kwee IL, Igarashi H, Suzuki Y. Aquaporin-4 functionality and VirchowRobin space water dynamics: physiological model for neurovascular coupling and glymphatic flow. Int. J. Mol. Sci. 2017;18:1798.
5. Conclusion The results of this study strongly support the anatomical substrate of FHR identified on a 7T system to be the GLE and its immediately adjacent interstitial space. The values of the GLE’s intrinsic contrast parameters, ρapp, T1, T2, show unique characteristics compatible with the structural and functional specifics of the GLE. The GLE is composed of snuggled endfeet of astrocytes with abundant AQP-4, the main water channel in brain. AQP-4 plays an essential role in the water dynamics of the brain inside the blood brain barrier. Since the unique combination of contrast parameters of FHR is strongly dependent on functionality of AQP-4, microscopic imaging of the GLE and analysis of its contrast parameters can be developed into an index of AQP-4 functionality at the GLE, and its possible role in neurological diseases. Acknowledgements This study is supported by grants from the Ministry of Education, Culture, Sports, Science, and Technology (Japan) and University of Niigata. The study was in part presented at the Society of Neuroscience 2015, 578.05/C23. References [1] Adams JG, Melhem ER. Clinical usefulness of T2-weighted fluid-attenuated inversion recovery MR imaging of the CNS. Am. J. Roentgenol. 1999;172:529–36. [2] Stuckey SL, Goh TD, Heffernan T, Rowan D. Hyperintensity in the subarachnoid space on FLAIR MRI. Am. J. Roentgenol. 2007;189:913–21. [3] Nakada T, Matsuzawa H, Kwee IL. High resolution imaging with high and ultrahigh-field MRI systems. NeuroReport 2008;19:7–13. [4] Nakada T, Matsuzawa H, Igarashi H, Fujii Y, Kwee IL. In vivo visualization of senile plaque like pathology in Alzheimer's Disease patients by MR microscopy on a 7T system. J. Neuroimaging 2008;18:125–9. [5] van Veluw SJ, Fracasso A, Visser F, Spliet WG, Luijten PR, Biessels GJ, et al. FLAIR images at 7 Tesla MRI highlight the ependyma and the outer layers of the cerebral cortex. NeuroImage 2015;104:100–9. [6] Rash JE, Yasumura T, Hudson CS, Agre P, Nielsen S. Direct immunogold labeling of aquaporin-4 in square arrays of astrocyte and ependymocyte plasma membranes in rat brain and spinal cord. Proc. Natl. Acad. Sci. U. S. A. 1998;95:11981–6. [7] Neely JD, Christensen BM, Nielsen S, Agre P. Heterotetrameric composition of aquaporin-4 water channels. Biochemistry 1999;38:11156–63. [8] Nakada T. Neuroscience of water molecules: a salute to Professor Linus Carl Pauling. Cytotechnology 2009;59:145–52. [9] MacQueen JB. Some methods for classification and analysis of multivariate observations. Proc Fifth Berkeley Symp on Math Statist and Prob. 1. 1967. p. 281297. [10] Vijayalakshmi P, Selvamani K, Geetha M. Segmentation of brain MRI using k-means clustering algorithm. Int. J. Engl. Trends Technol. 2011;3:113–5. [11] Hyvärinen A. Fast and robust fixed-point algorithms for independent component analysis. IEEE Trans. Neural Netw. 1999;10:626–34. [12] http://research.ics.aalto.fi/ica/fastica/. [13] McKeown MJ, Jung TP, Makeig S, Brown G, Kindermann SS, Lee TW, et al. Spatially independent activity patterns in functional MRI data during the stroop colornaming task. Proc. Natl. Acad. Sci. U. S. A. 1998;95:803–10.
145