1 INTRODUCTION

Ultrahigh-field MRI of whole post mortem brains permit long imaging sessions without motion artifacts, which brings about enhanced image quality at a high spatial resolution. The obtained data are complementary to histology and pathology studies and can be useful for tissue characterization, sectioning, 3D reconstruction of sectioned tissue data, and for diagnosis and investigation of neurodegenerative diseases.1-6

Ex vivo brain samples are preserved by perfusion and/or immersion in a fixative agent to slow down tissue decomposition.7-9 To optimize MRI, fixatives must be MR-compatible, suitable for tissue storage, and yield reproducible measurement results. To be MR-compatible means to have dielectric properties that ensure a homogeneous B1 field, a magnetic susceptibility matching the tissue to improve B0 homogeneity, and preferably a singulet 1H-NMR spectra to prevent chemical-shift artifacts.

Dielectric properties (electric conductivity, σ, measured in Siemens per meter [S/m]; and permittivity, ϵ, a unit-less measure) of the fixative can influence ex vivo whole human-brain MRI image quality at 9.4 T, considering their large size.10 Using the same RF coils and clinical MRI protocols as those used in vivo, choosing in vivo–like properties may represent a viable way to maintain image homogeneity and brain-tissue contrast. Analytical solutions of Maxwell’s equations governing RF-field transmission inside a sphere demonstrate how spatial field focusing and dielectric resonance can affect B1 field homogeneity at high frequencies due to the dielectric properties of the sample.11 Radiofrequency penetration, power deposition, and the spatial shape of the electromagnetic fields therefore directly influence MRI quality. Such effects may be particularly evident in T1-weighted and T2-weighted MRI acquisitions using 180º pulses, when spin inversion and echo refocusing may fail, leading to strong contrast differences between areas where different flip angles are played out.

The most well-established chemical fixation agent is formaldehyde (formalin), although currently there is an ongoing quest for alternative fixative agents, in considering its recent classification as a carcinogen (type 1B in Europe and group 1 of International Agency for Research on Cancer).12 Even at low concentrations, formalin can achieve its biocide action and stabilize tissue structure by cross-linking proteins.7-9 The rate of penetration of formalin can be quantified by the Medawar coefficient8 and can be facilitated through the use of phosphate-buffered formalin.9 This motivates its use for ex vivo MRI, but the dielectric properties of standardized formalin are not sufficient to yield good image quality at 9.4 T.10 Due to changes in water mobility and chemical exchange,13, 14 formalin causes a shortening of relaxation times.15, 16 This can be reversed by washing out the formalin and replacing it with PBS, which has been proposed as a method to increase the SNR.17 Nevertheless, washing out is not complete after 24 h for whole human brains.18 Phosphate-buffered saline (ϵ = 76.4, σ = 1.73 S/m) also causes B1 inhomogeneity at 9.4 T.10 Because fixative penetration and tissue fixation at all depths is a time-consuming process for large specimens like whole post mortem human brains, knowledge regarding these processes is required to ensure stable conditions and reproducible MRI measurements.

Here we investigated the possibility of developing formalin-based fixatives with additives to improve the dielectric properties, while assuring adequate tissue penetration, good fixation quality, and known MR properties. Additives were selected to achieve appropriate (1) dielectric properties that enable sufficient RF penetration and B1 homogeneity; (2) magnetic susceptibility that matches the tissue to obtain good B0 homogeneity and avoid susceptibility difference–related effects; and (3) chemical shift to ensure absence of offset artifacts. Additives like salt, polyvinylpyrrolidone (PVP), and low percentages of ethanol were used to decrease permittivity while salts, like NaCl and KCl, were used to adjust tonicity and conductivity. One PVP-based compound is known to have slight fixative properties.19 It also has interesting properties for MRI, due to its complex pattern in 1H-NMR spectra that avoids chemical-shift artifacts even at high temperatures, and its peculiar dielectric properties that allows tuning permittivity.20 The B1-field homogeneity for MRI at 9.4 T was assessed in solutions and in pig brain tissue samples during 0.5–35 days of immersion fixation. We also investigated the MR properties (T1, , R2, , and magnetic susceptibility [QSM]) at different tissue depths in white and gray matter of pig brain samples. The kinetics were fitted using exponential functions similar to previous studies.15, 21 The immersion time required to reach maximum values was used to estimate the Medawar coefficient for fixative penetration.8 The impact of embedding samples in Fluoroinert or PBS was also evaluated.17

2 METHODS 2.1 Preparation of fixatives and assessment of chemical shift–offset artifacts

Ready-to-use formalin (Roti-Histofix 4%, phosphate-buffered formaldehyde solution; pH 7; Carl Roth, Karlsruhe, Germany), polyvinylpyrrolidone (PVP; average molecular weight = 40 000 g/mol), ethanol (absolute for high-performance liquid chromatography; ≥ 99.8%), sodium (NaCl; BioXtra; ≥ 99.5% -), and potassium chloride (KCl, anhydrous, American Chemical Society reagent; ≥ 99.5%) (Sigma-Aldrich Chemie, Merck [Taufkirchen, Germany]) were used.

The 1H-NMR spectra of additives were generated using an online NMR predictor tool on the nmrdb.org website (Supporting Information Figure S1). Chemical shift–offset artifacts of ethanol (1%, 5%, 10%, 20%, and 40% [vol/vol]) in water with 0.9% NaCl were measured in five vials positioned inside a 2.8-L container using a holder. The outer container was filled with water, salt (0.9% [wt/vol]), and Dotarem (0.65 mM) as a T1 modifier. The phantom was scanned at 9.4 T using single-slice gradient-echo MRI with different readout bandwidths (90, 120, 240, and 600 Hz/pixel). Due to visual inspection and the criteria of ghost/signal ratio less than 1/40, an acceptable allowance of 5% was determined for the ethanol concentration (Supporting Information Figure S2).

2.2 Measurements of dielectric properties

The DAK (Dielectric Assessment Kit) probe DAK-12 (Schmid & Partner Engineering [SPEAG], Zürich, Switzerland) connected to a Vector Network Analyzer (Keysight ENA 5071C; Agilent, Santa Clara, CA) was used to measure ϵ and σ within the range of 100 MHz to 4.5 GHz at room temperature.

Formalin samples (250 ml) prepared with different combinations of NaCl (0%, 0.01%, 0.04%, 0.08% [wt/vol]) and PVP (0%, 2%, 5%, 10%, 20% [wt/vol]) concentrations (Figure 1) or with NaCl and ethanol (1%, 3%, 5% [vol/vol]) (Supporting Information Figure S3) were prepared and measured with DAK. The dependence of permittivity and conductivity on the concentrations of the additives was fitted with a polynomial in MATLAB (MathWorks, Natick, MA).

Conductivity (S/m) and permittivity (unit-less) for neutral buffered formalin using increasing amounts of PVP and NaCl (wt/vol percent) as additives at 400 MHz. The change in conductivity could be described with a third-order polynomial dependence on polyvinylpyrrolidone (PVP), and a second on NaCl concentrations, with a RMS error of 0.016 and a R2 equal to 0.989. Likewise, permittivity showed a third-order polynomial dependence on the PVP and the NaCl concentrations, with RMS error = 0.37 and R2 equal to 0.997

Depending on our selection criteria (see Section 3), four recipes for the fixative agents were defined, prepared four times at a 1-L volume to assess reproducibility and measured with DAK:

Fix01: Formalin + 0.04% NaCl + 5% PVP + 5% ethanol

Fix02: Formalin + 0.9% NaCl + 5% PVP + 5% ethanol

Fix03: Formalin + 0.9% NaCl

Fix04: Formalin + 0.8% NaCl + 0.2% KCl (corresponding to 140 and 2.7 mM, respectively).

For MRI, each fixative solution was prepared to be measured in an elliptic container (polycarbonate, 2.5 L) and in one of the chambers used for tissue samples.

2.3 Fixation and embedding of tissue samples

Two measurement-series were performed with pig brain hemispheres from the local slaughter center (post mortem interval < 5 h). Samples were kept at room temperature at all times and were placed in a four-chamber elliptic container (each chamber ~0.65 L, 3D-printed using polylactic acid with polycarbonate lids), after removal of dura and arachnoid to improve fixation and avoid artifacts from superficial blood vessels. Each chamber was filled with an embedding media and the samples immobilized by sterile gauze soaked in the same liquid. Air bubbles were removed using a desiccator.

In the first measurement series, four hemispheres, one per fixative, were embedded in their own fixative. The four samples were scanned at 9.4 T simultaneously at 12 h, 1, 2, 3, 4, 13, 19, 28, and 35 days of immersion fixation. Next, three hemispheres fixed with the same fixative were placed in three of the chambers with fixative embedder, with the fixative alone in the fourth. These experiments were performed after 38, 40, 41, and 42 days of fixation for Fix04, Fix03, Fix01 and Fix02, respectively, to check reproducibility.

In the second measurement series, four hemispheres (one for each fixative) were immersed for 28 days and then embedded in Fluorinert (FC-770) for MRI. Next, after 2 months of fixation, washing by immersion in PBS, replaced after 4, 8 and 12 h, was performed before final embedding in PBS. The MRI was repeated after keeping the samples in the chambers at room temperature for 1 month of PBS storage.

2.4 Magnetic resonance imaging measurements

The MRI data were acquired at 9.4 T using a whole-body system (Siemens, Erlangen, Germany) with a 16-channel-transmit/31-element receive array.22 B1-maps were obtained from actual flip-angle images23-25 (TE/TR1/TR2 = 7/20/100 ms; flip angle = 60°, voxel size = 3.1 × 3.1 × 5.0 mm3, FOV = 200 × 162.6 × 36 mm3), scaled with the transmitter voltage and a conversion factor to achieve the unit nT/V.

Parametric T1 maps were obtained from actual flip-angle image–corrected MP2RAGE measurements26; (TI1/TI2 = 900/3500 ms; TR = 6 ms; TE = 2.3 ms; volume TR = 9 s; flip angle = 4/6°; voxel size = 0.8 × 0.8 × 0.8 mm3), also used to generate background-free MP2RAGE beta images.27 The and QSM were obtained from a monopolar multi-echo gradient echo (TR = 34 ms; TE = 6.03, 12, 18, 24, 30 ms; flip angle = 15°; voxel size = 0.4 × 0.4 × 0.4 mm3; FOV = 204 × 165.8 × 46 mm3). The maps were obtained using nonlinear least-squares fitting, and QSM images were generated from the third gradient echo after Laplacian unwrapping RESHARP28 (kernel, 1.2 mm; Tikhonov regularization, 10−12) and dipole-inversion with iLSQR (https://people.eecs.berkeley.edu/~chunlei.liu/software.html; v. 3.0, 201905). Phase referencing was performed by setting the median value across all four samples to zero.

The R2 maps were obtained from single-slice Carr-Purcell-Meiboom-Gill spin-echo images; (TR = 3 s; 32 echoes in steps of 7.2 ms; TE1 = 7.2; TE32 = 227.6 ms; voxel size = 0.47 × 0.47 × 2 mm³; FOV = 192 × 192 × 2 mm3) followed by analysis using the extended phase graph model, while taking into account the actual B1 distribution as described previously.29 The maps were generated by subtracting the R2 from the maps. Based on the B1-field maps, the T1 and R2 measurements were repeated twice, with different settings of the reference voltage to achieve the nominal flip angle in the center of each brain sample (200 V for Fix02/03/04; 133 V for Fix01).

2.5 Data analyses

Data analysis was performed using MATLAB commands after affine co-registration and nearest-neighbor resampling to match the image at 24 h in SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). Whole brain masks were obtained from the MP2RAGE contrast image using activecontour and imclose. The function distancemap from https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSL yielded the tissue depth from the whole brain mask. Automatic whole brain gray-matter (GM)/white-matter (WM) segmentation was performed on /T1-ratio maps using histeq and thresholding at 0.98. The GM and WM masks were also carefully outlined on the most central 10 slices using activecontour, followed by manual correction if deemed necessary.

The MR parameters were extracted from WM and GM voxels to generate median values. Nonlinear curve fittings of the time evolutions from 10 slices between 0.5 and 35 days were made with cftool. A bi-exponential function was fitted for T1, mono-exponential for QSM, and a sum of a mono-exponential decay and mono-exponential recovery for (Table 2), in line with previous studies.15, 21 Median values for the depth-dependent B1 values and DMAX, the day at which the maximum was reached determined from pixelwise fits, were extracted from whole-brain GM, WM masks whenever ≥ 50 voxels were available within a certain depth bin, going from 0.5 to 15.5 mm with a bin width of 1 mm. The DMAX was only extracted if the goodness-of-fit in terms of the adjusted regression coefficient  > 0.8, and if DMAX < 25 days. The Medawar coefficient K was obtained from , where t = DMAX * 24 is the median DMAX expressed in hours.

The 95% confidence intervals for the prediction error of the exponential fits (predint) were generated and used to verify significant deviations of the MR properties at day 38–42, and for different embedding media. For B1, the coefficient of variation across three samples on day 38–42 was used to evaluate significance, as no variation with immersion time was observed. Statistical analyses using three-way analysis of variance (ANOVA) comparing tissue (WM-GM), fixatives (Fix01–04), and immersion time (0.5–35 days, treated as a continuous variable), and two-way ANOVA to evaluate WM-GM contrast across fixatives and immersion time, were performed using anovan, requiring p < 0.05 for significance. For the change in R2 (day 1 vs. 28) and (day 4 vs. 28), two-tailed t-tests were performed.

3 RESULTS 3.1 Dielectric properties of formalin-based solutions and selection of tissue fixatives

Both conductivity and permittivity of formalin decreased with increasing PVP concentrations, occurring at a higher rate in the presence of NaCl (Figure 1). The change in conductivity could be described with a third-order polynomial dependence on PVP, and a second on NaCl concentrations, with a RMS error of 0.016 and R2 = 0.989. Likewise, permittivity showed a third-order polynomial dependence on the PVP and the NaCl concentrations, with RMS error = 0.37 and R2 equal to 0.997. Effects of adding different amounts of NaCl and ethanol are shown in Supporting Information Figure S3. By adding either NaCl, PVP, or ethanol to the formalin, the permittivity decreased, with the greatest effect obtained by ethanol (Table 1). While adding NaCl led to an increase in conductivity, ethanol slightly reduced it. The addition of all three components did not lead to a linear decrease in permittivity.

TABLE 1. Effect of additives on the dielectric properties of formalin (Roti Histofix 4%) Chemicals Δσ (S/m) Δε Formalin (reference) 0 0 Formalin + 0.9% NaCl +1.33 −3.5 Formalin + 5% PVP −0.06 −4.4 Formalin + 5% ethanol −0.13 −5.3 Formalin + 0.9% NaCl + 5% PVP + 5% ethanol (Fix02) +0.83 −8 Note The change in dielectric properties owing to all additives used for Fix02 is not simply a linear combination of the differences caused by adding each single material. Abbreviations: Δσ (S/m), conductivity difference; Δε, permittivity difference.

Additives were selected to achieve near to brain-equivalent dielectric properties: 0.04% NaCl, 5% PVP, and 5% ethanol (Fix01). Besides further reducing permittivity, ethanol facilitated tissue penetration. For Fix02 we used the same additives but increased the NaCl concentration to isotonic levels (0.9%), to minimize the risk of osmosis and cellular swelling. Fix03 was defined to study effects in absence of PVP and ethanol. Fix04 contained the exact physiological concentration of the two most common salts in tissue cells and was evaluated for consistency with previous measurements.10 The conductivity at 400 MHz and room temperature was 0.60 ± 0.01, 1.55 ± 0.08, 1.98 ± 0.02, and 1.89 ± 0.01 S/m for Fix01, Fix02, Fix03 and Fix04, respectively, and the permittivity was 71.2 ± 0.24, 70.3 ± 0.16, 73.8 ± 0.05, and 74.0 ± 0.23, respectively. The values obtained at other MRI operating frequencies are listed in Supporting Information Table S1.

3.2 B1-transmit field mapping of the fixatives

The B1 field of each fixative was measured in the absence of tissue in a 2.5-L container (Figure 2, upper row) and in one of the four chambers, while the remaining chambers contained brain samples immersed in the corresponding fixative (Figure 2, lower row). For Fix01, a nonuniform B1 field with a central hot spot enclosed by a region where the B1 field was about 90% weaker was observed (Supporting Information Figure S4). The other fixatives yielded a more homogeneous B1 field. The corresponding histogram (Figure 3A) was wider for Fix01, encompassing central voxels with high efficiency. The remaining fixatives more closely adhered to a Gaussian distribution. Ideally, to achieve the best performance for MRI, the fixatives should yield histograms with a narrow peak at high B1. The SD of the B1 field across the whole volume was 10.36, 6.38, 6.80, and 7.36 nT/V, respectively, for Fix01–Fix04. In the smaller container, histograms (Figure 3B) suggest that besides the difference in size and loading by tissue samples, the material of the container itself also affected the B1 field, with SDs of 24.81, 8.59, 6.29, and 7.93 nT/V, respectively, for Fix01–Fix04.

The B1 field at 400 MHz of each fixative solution was measured in an elliptical 2.5-L container in polycarbonate designed for measuring whole post mortem human-brain samples (upper row). The B1 pattern and dielectric properties of Fix01 are close to the ones encountered in vivo at 9.4 T. The measurements were repeated for each fixative solution in a four-chamber elliptical container, which was 3D-printed using polylactic acid and had polycarbonate lids (lower row). The upper-left quarter of the ellipsoid was filled with the fixative alone, whereas the other three contained the same fixative in the presence of pig-brain samples that had been preserved in the fixative for 38 (Fix04), 39 (Fix03), 41 (Fix01), or 42 days (Fix02). The B1 maps suggest that the containers themselves and the presence of tissue can affect B1-field uniformity. The (mean) dielectric properties of each fixative at 400 MHz and room temperature (± SD from four repetitions) were as follows: 0.60 ± 0.01/1.55 ± 0.08/1.98 ± 0.02/1.89 ± 0.01 S/m for the conductivity of Fix01/02/03/04, respectively, and 71.2 ± 0.24/70.3 ± 0.16/73.8 ± 0.05/74.0 ± 0.23 for the permittivity (see Supporting Information Table S1) Measured B1 values in the absence and presence of brain tissue. Shown are histograms for the fixative solutions measured in a 2.5-L elliptic container designed for human whole-brain samples (A) and for one of the four chambers of an elliptical container that contains unloaded fixative (B) shown in Figure 2. Fix01 has a wider histogram, while the values of the remaining fixatives more closely adhere to a Gaussian distribution. The differences in B1 between the fixatives become more pronounced in the tissue samples (B) but remain stable during 0.5–35 days of immersion fixation both in gray matter (GM) (C) and in white matter (WM) (D). The influence of tissue depth (E,F) is prominent and exceeds the SD across 0.5–35 days of immersion fixation, shown as bars. The B1 values in WM voxels (D,F) are higher than in the GM (C,E), especially at earlier immersion times, at increasing tissue depths and in the presence of PVP (Fix01 and Fix02) 3.3 B1-transmit field mapping of brain-tissue samples in different fixatives

Comparison of B1 maps obtained in absence and in presence of brain tissue demonstrates that the sample itself modifies B1 (Figure 2).

A three-way ANOVA analysis of extracted B1 values revealed significant main effects of tissue, fixative, and duration of fixation (F[56,1] ≥ 16, p ≤ 0.0002) and a significant tissue-by-fixative interaction (F[56,3] = 4.36, p = 0.0079). The highest B1 values were found in WM with Fix01 at early and late time points (Figure 3C,D). At increasing tissue depths, prominent effects that outsized the effect of immersion time were observed (Figure 3E,F).

3.4 Effect of embedding media on B1 in fixed brain tissue

The advantage of using a fixative with optimized dielectric properties in terms of obtaining high B1 within the tissue was diminished when the surrounding fixative was replaced by other embedding media, but did not completely disappear (Figure 4). The B1 in GM measured with fixative embedding was 40% higher for Fix01. With Fluorinert as an embedder, this number reduced to 5%–20%, and with PBS, Fix01 was still 15%–25% better. Likewise, in WM, Fix01 had 2%–15% higher B1 in Fluorinert and 20%–25% in PBS. The embedding-induced change in B1 of Fix02 exceeded the coefficient of variation on day 38–42 in WM, whereas in GM the changes were just within. On the contrary, in the PVP-free fixatives, higher B1 values were observed with both alternative embedding media. No effect of 1-month storage in PBS was observed.

Effect of embedding media on B1 values measured in GM (A) and WM (B) voxels. Pig-brain samples were immersed for 28 days in each fixative before measurements while embedded in its own fixative (blue, Fix) or in proton-free media (orange, Fluor). After 2 months of fixation the samples measured in Fluorinert were washed with phosphate-buffered saline (PBS) for 24 h before embedding in PBS and MRI (yellow, PBS2). After 1 month of storage in PBS, MRI was performed again (PBS3, violet). Significance (p < 0.05 indicated by a star) was assessed based on the coefficient of variation across three samples measured in each fixative on day 38–42

3.5 Magnetic resonance properties in fixed brain tissue

Example maps of tissue , T1, and QSM during immersion are shown in Figure 5 together with the exponential fits of the time evolution in GM/WM. The (T1-) QSM values diminished (bi-)exponentially with time, whereas showed an increase followed by a decrease (Table 2). The reproducibility data at day 38–42 were not always within the 95% confidence intervals of the fits (Supporting Information Figure S5), especially for Fix03, which had several remaining air bubbles. For QSM and , effects due to fiber orientation in WM could also have contributed to the variance.

Magnetic resonance properties of pig-brain samples measured at different timepoints during immersion fixation using four fixatives. Maps and time courses are shown for T1 (A,B), (C,D), and QSM (E,F). Maps show sagittal slices located at the center of each hemisphere at 12 h, 4 days, 13 days, and 35 days for Fix01-Fix04 (left to right column). Time courses of extracted median values for GM (A,C,E) and WM (B,D,F) voxels were fitted with exponential functions between 12 h and day 35, with coefficients listed in Table 2. The mean and SD in three hemispheres per fixative on day 38–42 are shown for comparison, while the 95% confidence intervals are shown in Supporting Information Figure S5 TABLE 2. Exponential functions describing the change in MR properties (T1, , and QSM) in GM and WM tissue voxels during immersion fixation with four different fixative solutions Fixative, tissue [ms] a [d−1] b [d−1] [ms] [s−1] a [d−1] b [d−1] [s−1] QSMa [ppb] a [d−1] QSMb [ppb] Fix01, GM 379 0.53 0.005 1151 0.977 44.6 0.23 0.36 48.2 0.824 4.50 0.31 0.70 0.751 WM 426 0.48 0.002 964 0.967 35.5 0.10 0.36 64.1 0.659 3.59 0.60 −10.30 0.035 Fix02, GM 441 0.60 0.005 1072 0.972 44.2 0.26 0.42 45.4 0.917 7.69 0.86 3.06 0.888 WM 561 0.75 0.004 933 0.988 37.4 0.11 0.42 63.4 0.977 16.86 0.86 −8.31 0.936 Fix03, GM 243 0.91 0.004 1261 0.981 52.9 0.20 0.28 39.5 0.994 5.65 0.50 1.30 0.893 WM 544 1.83 0.002 1032 0.991 43.2 0.22 0.59 53.7 0.985 14.30 1.49 −9.17 0.634 Fix04, GM 232 0.47 0.004 1263 0.964 51.3 0.17 0.23 39.9 0.976 10.95 0.68 0.02 0.803 WM 390 0.88 0.002 1036 0.954 46.2 0.15 0.38 55.2 0.993 13.86 1.74 −13.01 0.672 Note The exponential rate constants are expressed in units per day [d−1]. For a graphical view, see Figure 5 and Supporting Information Figure S5.

The three-way ANOVA showed significant main effects of tissue type and immersion time (F[56,1] ≥ 6.66; p ≤ 0.0125). For QSM there was also a significant difference between fixatives (F[56,1] = 8.09; p = 0.0001) and a significant tissue-by-fixative interaction effect (F[56,3] = 3.31; p = 0.026). Although the GM in Fix02 was the most paramagnetic, the WM in Fix04 was the most diamagnetic. The two-way ANOVA showed a significantly different QSM difference between GM-WM (F[28,3] = 7.66, p = 0.0007), being greater in the absence of PVP. The GM-WM contrast in changed significantly with immersion time, reaching the highest difference in Fix02 (26 s−1) on day 4 (F[28,1] = 7.52; p = 0.010), although the effect of fixative was not significant (F[28,3] = 0.33; p = 0.805). The T1 contrast was significantly different between fixatives (F[28,3] = 13.61; p < 0.0001) and across time (F[28,1] = 18.32; p = 00002), being greater in the abse