key: cord-0265056-r64gvvdl authors: Chan, Kwok-Shing; Hédouin, Renaud; Mollink, Jeroen; Schulz, Jenni; Walsum, Anne-Marie van Cappellen van; Marques, José P. title: Studying magnetic susceptibility, microstructural compartmentalisation and chemical exchange in a formalin-fixed ex vivo human brain specimen date: 2021-08-01 journal: bioRxiv DOI: 10.1101/2021.07.30.454493 sha: 1bc61cd2cff1e407ed7a4c6a404b6ddfc8c4042c doc_id: 265056 cord_uid: r64gvvdl Purpose Ex vivo imaging is a preferable method to study the biophysical mechanism of white matter orientation-dependent signal phase evolution. Yet, how formalin fixation, commonly used for tissue preservation, affects the phase measurement is not fully known. We, therefore, study the impacts of formalin fixation on magnetic susceptibility, microstructural compartmentalisation and chemical exchange measurement on human brain tissue. Methods A formalin-fixed, post-mortem human brain specimen was scanned with multiple orientations with respect to the main magnetic field direction for robust bulk magnetic susceptibility measurement with conventional quantitative susceptibility imaging models. Homogeneous white matter tissues were subsequently excised from the whole-brain specimen and scanned in multiple rotations on an MRI scanner to measure the anisotropic magnetic susceptibility and microstructure-related contributions in the signal phase. Electron microscopy was used to validate the MRI findings. Results The bulk isotropic magnetic susceptibility of ex vivo whole-brain imaging is comparable to in vivo imaging, with noticeable enhanced non-susceptibility contributions. The excised specimen experiment reveals that anisotropic magnetic susceptibility and compartmentalisation phase effect were considerably reduced in formalin-fixed white matter tissue. Conclusions Despite formalin-fixed white matter tissue has comparable bulk isotropic magnetic susceptibility to those measured via in vivo imaging, its orientation-dependent components in the signal phase related to the tissue microstructure is substantially weaker, making it less favourable in white matter microstructure studies using phase imaging. Quantitative susceptibility mapping (QSM) is a physics-driven method to the study magnetic 60 properties of biological tissues (1). Some features that differentiated it from conventional MR 61 relaxometry include the field strength independence of the derived maps, relying on spatial 62 deconvolution and its ability to distinguish paramagnetic and diamagnetic substances since 63 they produce opposite contrasts. QSM is commonly performed on gradient echo phase data 64 owing to its direct relationship to magnetic field variations (2,3). 65 66 One major QSM research challenge is to understand the mechanism of phase evolution in 67 white matter (WM) (4,5). In deep grey matter (GM), strong correlations between QSM and iron 68 concentration have been demonstrated (6). Yet, in WM, the abundance of diamagnetic myelin 69 (relative to water) would have suggested a strong QSM contrast relative to cerebrospinal fluid 70 (CSF) (4,5). The lack of this strong contrast has been attributed to various biophysical 71 phenomena (7,8). The lipid-rich myelin bilayer sheath encapsulating the highly-ordered axons 72 in WM results in anisotropic susceptibility (7-12). Additionally, water protons exist in various 73 microstructural environments (13) , namely myelin water, intra-axonal water and extra-axonal 74 water, which can have different signal decay rates and frequency shifts depending on the 75 composition of the tissue and the fibre orientation with respect to the main magnetic field (B0) 76 (14-17), which also make the signal phase not representing the average magnetic field in a 77 voxel. The chemical exchange of protons between macromolecules and water can also 78 introduce a further frequency shift of the MR signal (18, 19) . Disentangling the origins of WM 79 phase contrast can improve our understanding of QSM and provide new means to account for 80 their impact in QSM. 81 82 As WM phase contrast is orientation-dependent, studying its properties requires data acquired 83 with different orientations to B0. Subject compliance limits the range of angles that can be 84 obtained in vivo (unnatural posture inside the scanner). Experiments with ex vivo samples, on 85 the other hand, do not suffer from this limitation, allowing long scanning sessions without data 86 degradation caused by motion and for histology to be performed as a means of validation of 87 any microstructural findings (20,21). One shortcoming of ex vivo experiments is that MR 88 measured parameters in tissues undergone formalin fixation (a common practice to preserve 89 human post-mortem tissue) show substantial differences to those found in vivo. Those 90 differences have been seen both on single-and multi-compartment relaxometry, and diffusion-91 weighted imaging (22, 23 ). Yet, a previous study showed that the bulk magnetic susceptibility 92 of brain tissues measured by QSM did not change significantly between in vivo and ex vivo 93 4 Correspondence to Kwok-Shing Chan conditions, and also during a 6-week fixation period (24). This finding is in agreement with the 94 experiment results when studying the microstructural effect in phase imaging (25) , where 95 comparable bulk magnetic susceptibilities were observed between fresh and fixed rat optic 96 nerves but the origins of the susceptibility contrast are different. 97 In this study, we investigate the magnetic susceptibility, compartmentalisation and chemical 99 shift effects on the MR phase using a formalin-fixed, post-mortem human brain specimen for 100 WM phase-contrast mechanism studies at 3T, providing comprehensive insights for the use 101 of fixed tissue in future QSM methodology studies. We performed multiple orientation 102 experiments in both whole-brain and excised tissue samples (26). This enabled us to obtain 103 both traditional QSM maps and ground truth bulk magnetic susceptibility measurements of the 104 excised samples, as well as a separate estimation of microstructure compartmentalisation 105 information. The samples were then studied using electron microscopy (EM) to further 106 evaluate microstructural correlates between MR and histology. 107 108 5 Correspondence to Kwok-Shing Chan The study is divided into 3 parts: 110 1) a post-mortem, formalin-fixed human brain specimen was scanned on an MRI scanner in 111 multiple orientations with respect to B0 for robust magnetic susceptibility measurements; 112 2) homogeneous WM specimens were excised from the whole-brain specimen, embedded in 113 agar and subsequently scanned again in various orientations in respect of B0 to measure their 114 magnetic susceptibility and microstructure-induced field, similar to the experiment conducted 115 by (26); 116 3) the WM specimens used in the second part of the study were imaged by 3D EM, allowing 117 MRI data to be compared to histology. 118 119 Tissue processing 120 A post-mortem human brain specimen from a deceased male (aged 78 years old) with no 121 history of neurological disorder (cause of death: myocardial infarction) was used for this 122 research in accordance with the local ethics committee and the Anatomy Department of 123 Radboud University Medical Center (Radboudumc, Nijmegen, the Netherlands). The brain 124 specimen was immersed in 10% formalin for tissue fixation after being extracted from the skull. 125 126 After one month of fixation, the specimen was scanned on an MRI scanner (imaging details in 127 section 2.2) to obtain the brain morphology for creating a tailor-made holder. The holder was 128 made of a stack of 35 4-mm thick, 3D-printed plastic plates having space with the same shape 129 of the specimen in the centre, covering most of the brain (see Figure 1 ) where the specimen 130 can be fitted tightly inside the holder and a surrounding spherical container. Each plate has a 131 grid layout with 4 mm x 4 mm elements, providing landmarks in MRI images for planning and 132 guidance of tissue excision for the validation experiment in the second MRI session. to the cylindrical axis. Imaging was performed about 1 day after the excised specimens were 142 acquired and 5 days after the whole brain scan. Correspondence to Kwok-Shing Chan 144 EM was utilised to provide an additional reference to understand and explain the MRI findings. 145 Two days after the second MRI session, the WM specimens were sectioned to 100 m on a 146 vibratome (VT1000S, Leica Biosystems, Nussloch, Germany) before being immersed in 2.5% 147 glutaraldehyde in 0.1M sodium cacodylate buffer for overnight incubation at 4°C. The 148 specimens were then transferred to 0.25% glutaraldehyde in 0.1M sodium cacodylate buffer 149 for storage at 4°C, and then delivered to the EM facility at the University of Oxford for imaging. 150 The workflow of this study is summarised in Figure 1 164 Data acquisition 165 The study was approved by the local ethics committee. All MRI data were acquired on a 3T 166 scanner (Prisma, Siemens, Erlangen, Germany) at room temperature (20°C) using a 64-167 channel array head/neck coil (with only 48 head channels were enabled). The experiment 168 consisted of two imaging sessions: the first session was conducted on the whole-brain 169 specimen and the second session was conducted on the excised brain tissues. The following 170 protocol was used for the first session: 171 (1) MP2RAGE adopted to sensitise for T1 values between 250 ms and 1000 ms, 1 mm 172 isotropic resolution, TI1/TI2/TR=311/1600/3000 ms, flip angle ( ) #1/#2 = 4°/6°, Tacq 173 = 5 min; 174 (2) 2D spin-echo EPI DWI, 1.6 mm isotropic, TR/TE=15241/77.6 ms, 2-shell 175 (b=0/1250/2500 s/mm 2 , 17/120/120 measurements with 7 b=0 measurements 176 collected with reversed phase-encode blips for distortion correction), 20 repetitions, 177 In the second scanning session, the excised specimens were scanned with the following 184 protocol: 185 (1) MP2RAGE sequence with the same parameters as above; 186 (2) 2D spin-echo EPI DWI, 1 mm isotropic, TR/TE=15241/77.6 ms, 2-shell 187 (b=0/1250/2500 s/mm 2 , 17/120/120 measurements with 7 b=0 measurements), 9 188 repetitions, Tacq = 9 hours; 189 distortion and then linearly registered to a common space, independent of the experiment 201 orientations. R2* maps were computed using a closed-form solution (33). Field maps were 202 computed using SEGUE (34) spatial phase unwrapping with optimum-weighted echo 203 combination (35) and tissue field maps were computed using LBV (36) in SEPIA (37) . For the 204 whole-brain data, bulk isotropic magnetic susceptibility was derived using COSMOS (38). 205 Additionally, the QUASAR algorithm for multi-orientations was also applied to test if the bulk 206 isotropic magnetic susceptibility measurement improved when the field generated by non-207 susceptibility contributions (f ) were simultaneously estimated (39) Whole-brain imaging results are shown in Figure 2 . The R1 maps obtained from the first 278 session (5-month fixation) show faster relaxation rates than those from the pre-scan (1-month 279 fixation), with the contrast between WM and cortical GM being clearly reduced, and DGM 280 showing increased R1 (Figure 2A and 2B) . The COSMOS derived magnetic susceptibilities 281 are in reasonable agreement with previously published in vivo data (see supplementary Figure 282 S1), where opposite magnetic susceptibility between WM and GM can be observed (Figure 283 2C). However, the residual field of the COSMOS estimation shows a slowly-varying pattern 284 across the brain that cannot be explained by the isotropic dipole field ( Figure 2E ) and is 285 relatively stable across orientations (see Figure 2F ). This residual map shares similar contrast 286 and values with the QUASAR non-susceptibility contributions map ( Figure 2H ). Hence, not 287 surprisingly, the bulk magnetic susceptibilities derived from QUASAR ( Figure 2G Figure 3 shows the magnetic susceptibility of the excised specimens measured via the 299 external field on agar with their ROIs illustrated in the whole-brain R1 map. The mean i and 300 a are -1.17±9.18 ppb and 4.03±1.63 ppb across WM specimens. A relatively strong positive 301 i is found in the corticospinal tract specimen (CST; 19.17 ppb), which was found in retrospect 302 to be due to some DGM residual in one end of the excised sample. The coefficient A of sin 2 303 dependence reflecting the WM microstructure effect has a mean of 1.46±1.55 ppb with a mean 304 intercept B of -2.75±0.79 ppb in WM. However, the R 2 of the specimen residual field fitting 305 suggests that not all WM specimens fit the sin 2 function equally well, particularly for samples 306 obtained from the genu and splenium of the CC (CC7-CC9; R 2 ranging from 0.01 to 0.71). 307 Therefore, we focused on the 6 WM specimens obtained from the body of the CC (CC1-CC6) 308 in comparison to the whole-brain data in the linear regression analysis. 309 Strong linear relations were found in mean susceptibility estimated by COSMOS between the 318 excised specimens and the corresponding ROI in the whole-brain data (cross-session; 319 R 2 =0.603, Figure 4A ), between the i from external field measurement and the mean 320 COSMOS susceptibility in the whole-brain data (cross-session, cross-method; R 2 =0.783, 321 Figure 4B ), and between the i from external field measurement and the mean COSMOS 322 susceptibility on the excised specimens (cross-method; R 2 =0.925, Figure 4C ). All the slopes 323 of the linear regressions are close to 1, whereas the relatively large intercepts in Figure 4A Correspondence to Kwok-Shing Chan In this study, we examined the magnetic susceptibility and microstructural 350 compartmentalisation effect on MRI phase data on a formalin-fixed, post-mortem human brain 351 specimen. The bulk magnetic susceptibility of the whole-brain specimen shows comparable 352 contrast to those in the previous ex vivo studies (6,24), as well as to in vivo imaging: WM is 353 slightly diamagnetic, whereas cortical and deep GM are paramagnetic. Further investigation 354 reveals the residual fields of COSMOS have a gradient-like appearance varying from the 355 surface toward the centre of the specimen ( Figure 2E ), which is similar to the expected way 356 of how the solutions (fixative or water) diffused into the specimen. Since these residual fields 357 are relatively constant across different rotations, it is likely to be caused by the exchange effect 358 (17-19) . These fields were captured as the non-susceptibility contributions by QUASAR and 359 they did not have a significant impact on the bulk magnetic susceptibility measurement ( Figure 360 2C, 2G). This result is distinct from in vivo imaging results ( Figure S1 ), where the susceptibility 361 differences in WM are more noticeable, suggesting that the effect of (sub)cellular structure of 362 WM is considerably reduced and the sphere of Lorentz inclusion utilised in COSMOS is 363 already a good approximation on formalin-fixed tissue. 364 365 While the bulk susceptibility of the WM samples in this study is similar to in vivo imaging, the 366 residual field analysis inside the excised homogenous tissue confirmed that the microstructure 367 compartmental frequency (parameter A in Eq. 4) is notably weaker in our samples than in vivo 368 and reported by others. In a similar experiment (26), the amplitude of the microstructure 369 frequency of a fresh bovine optic nerve at 7T was -18.75 ppb, significantly larger in magnitude 370 and with an opposite sign to what we have obtained in our CC samples, 1.46 ppb. Additional 371 analysis was performed to consolidate this result (see supplementary Figure S2 ). A reduction 372 of the microstructural compartmentalisation effect had already been reported in the literature 373 when studying fresh vs fixed rat optic nerves (25) . One possible explanation is the structural 374 alteration of the myelin sheath in fixed tissues. In our 3D EM images, we observed myelin 375 sheath spitting and swelling in some of the myelinated axons, similar to the observation 376 reported in the previous study (41), and such phenomena appeared more frequently in larger 377 axons than small axons. Based on the general Lorentzian tensor approach (43), the increase 378 of the aqueous space of the myelin sheath can result in the amplitude reduction of the induced 379 frequency shifts inside the myelin sheath and the intra-axonal space. Microstructural 380 differences related to structures (bovine optic nerve vs human CC) and age-associated 381 demyelination (45,46), together with the tissue preparation methods can also contribute to the 382 Sensitivity of MRI resonance frequency to the 480 orientation of brain tissue microstructure In vivo visualization of myelin water 483 in brain by magnetic resonance Micro-compartment specific T2⁎ relaxation in the 485 brain Nonexponential T2* 487 decay in white matter Gradient echo based fiber orientation mapping using R2* and 490 frequency difference measurements Fiber orientation-dependent white matter contrast in gradient echo 493 Protein-induced water 1H MR 496 frequency shifts: Contributions from magnetic susceptibility and exchange effects Investigating lipids as a source of chemical 499 exchange-induced MRI frequency shifts In vivo histology of the myelin g-ratio with magnetic 502 resonance imaging Postmortem Validation Study Formalin tissue fixation biases myelin-506 sensitive MRI Characterization of Diffusion, Relaxometry, and Myelin Water Fraction Measurements at 3T Ex-vivo quantitative susceptibility mapping of 512 human brain hemispheres Magnetic susceptibility induced white matter MR signal 515 frequency shifts-experimental comparison between Lorentzian sphere and generalized 516 Lorentzian approaches Effects of white matter microstructure on phase and susceptibility 519 maps Decoding the microstructural properties of white 521 matter using realistic models Denoising of 524 diffusion MRI using random matrix theory How to correct susceptibility distortions in spin-527 echo echo-planar images: application to diffusion tensor imaging Advances in functional and structural MR 530 image analysis and implementation as FSL An integrated approach to correction for off-resonance 533 effects and subject movement in diffusion MR imaging ANTs similarity metric performance in brain image registration A modulated closed form solution for 539 quantitative susceptibility mapping--a thorough evaluation and comparison to iterative 540 methods based on edge prior knowledge SEGUE: a Speedy rEgion-Growing algorithm for Unwrapping 543 Estimated phase An 546 illustrated comparison of processing methods for MR phase imaging and QSM: combining 547 array coil signals and phase unwrapping Background field removal by solving the Laplacian 550 boundary value problem SEPIA-Susceptibility mapping pipeline tool for phase images Calculation of susceptibility 554 through multiple orientation sampling (COSMOS): A method for conditioning the inverse 555 problem from measured magnetic field map to susceptibility source image in MRI. Magnetic 556 resonance in medicine Quantitative susceptibility mapping (QSM) with an extended 558 physical model for MRI frequency contrast in the brain: a proof-of-concept of quantitative 559 susceptibility and residual (QUASAR) mapping A semi-automated approach 562 to dense segmentation of 3D white matter electron microscopy MRI artifacts in human brain tissue after 565 prolonged formalin storage User-guided 3D active contour segmentation 568 of anatomical structures: Significantly improved efficiency and reliability Lorentzian effects in magnetic susceptibility mapping of 571 anisotropic biological tissues The effect of realistic geometries 574 on the susceptibility-weighted MR signal in white matter A representative reference for MRI-based 577 human axon radius assessment using light microscopy Correspondence to Kwok-Shing Chan differences observed in this study, as all these factors modulate the relative water 383 concentration in the three WM compartments. 384 385 All six specimens obtained from the body of the CC have similar magnetic susceptibility 386 anisotropy, suggesting that they have similar MVF based on the HCM approximation (Eq. S25 387 in (17)), and this is supported by the EM analysis (0.278 & 0.257 between the two samples). 388The amplitude of the residual field inside the specimens is, on the other hand, subject to 389 various properties including MVF, AVF and the aggregate g-ratio of the sample (Eq. A14 in 390 (16)). Interestingly, the realistic geometry of the WM fibre also plays an important role in the 391 compartmental frequency shifts (27, 44) . This effect is clearly illustrated in the frequency 392 perturbation simulations in Figure 5 : not only the centres but also the FWHM of the extra-393 cellular frequency distribution of the two samples are different, despite the two specimens 394 having virtually identical MVF and AVF. The broader frequency spectrum of CC4 induces a 395 faster R2* decay in the extra-axonal space and the specimen also has a more disperse fibre 396 arrangement. These two factors together reduce the amplitude discrepancy between the slow 397 R2* (intra-and extra-axonal water) and the fast R2* (myelin water) compartments throughout 398 the echo time as well as the frequency difference between those compartments that could 399 result in a reduced compartmentalization effect.