Cases
Our study cohort consists of twenty-two post-mortem human brains that were included through the body donors’ program at the Radboud university medical center, Nijmegen, The Netherlands between 2015 and 2019. Patients were included when hypertension (according to national guidelines at that time) was reported in their medical records. Age-matched individuals were included as controls when no record of hypertension or use of antihypertensive medication was identified in their medical reports. Exclusion criteria were the presence of a brain tumor and/or metastases, and/or territorial infarctions and/or atrial fibrillation (based on medical history or when identified during post-mortem MRI) as the latter two may also result in thromboembolic MRI lesions that mimic MRI markers of SVD. All individuals signed informed consent to use their medical records for research purposes, autopsy and use of tissue. The study was approved by the Medical Ethics Review Committee region Arnhem-Nijmegen (Commissie Mensgebonden Onderzoek (CMO) region Arnhem-Nijmegen file No. 2017–3941).
Vascular risk factors in medical history
Presence of diabetes was based on either a reported diagnosis, or the use of antidiabetic medication in the medical records. Hypercholesterolemia was based on a history of statin use and/or report of hypercholesterolemia or elevated cholesterol levels in the medical records. Classification of vascular risk factors was done in accordance with national guidelines at the time of death. Smoking and/or alcohol use were reported as ever/never-smoking or drinking alcohol.
Tissue processing
After autopsy, brains were removed from the skull and fixed in ~ 8% formalin for at least 2 months before imaging. Prior to HF MRI scanning, the brainstem with cerebellum and the circle of Willis were removed. Because of space limitations of the HF coil, the brain was cut mid-sagittal and the left hemisphere was horizontally divided into a dorsal and a ventral part using refence landmarks like the corpus callosum to keep tissue processing comparable between individuals. These parts were subjected to HF 7 Tesla MRI to visualize radiological markers of SVD. After careful examination of HF MRI data, the ventral part of each brain was divided into a slab of approximately 10 cm length along the horizontal plane (Fig. 1). For (immuno-)histopathology these slabs were cut along the coronal plane into blocks of approximately 2 × 2 × 0.5 cm containing periventricular white matter.
Post-mortem MRI
Before scanning, the specimens were removed from the formalin solution and washed in tap water for at least 24 h. Next, the ventral part of the left hemisphere was placed in a plastic bag filled with Fluorinert (3 M, FC-3283, Maplewood, MN, USA), a proton-free liquid. Air bubbles were removed by hand and by ultrasonic bath (Bransonic 221, Danbury, CT, USA). Specimens were scanned at room temperature in a Bruker 7 Tesla Clinscan MRI system (Bruker Biospin, Ettlingen, Germany) interfaced with a Siemens Syngo VB15 console. Scans were acquired with a T1-weighted sequence at a resolution of 400 × 400 × 400 µm (repetition time (TR) = 20 ms, echo time (TE) = 1 ms, 1 average); T2*-weighted sequence at a resolution of 400 × 400 × 400 µm (TR = 20 ms, TE = 13 ms, 1 average); T2-weighted sequence at a resolution of 400 × 400 × 400 µm (TR = 10,780 ms, TE = 25 ms, 2 averages); FLAIR sequence at a resolution of 500 × 500 × 500 µm (TR = 8200 ms, TE = 39 ms, 2 averages).
Post-mortem MRI scans were acquired with a mean of 11 months after death (standard deviation (SD) 11.9 months), MRI acquisition time was not different between individuals with hypertension and age-matched controls (p = 0.756). All acquired MRI images were displayed using MANGO (version 4.1; Multi-image Analysis GUI); Research Imaging Institute, University of Texas health science center, TX, USA; www.ric.uthscsa.edu/mango) [15]. Periventricular WMH severity was visually evaluated on MRI FLAIR sequence following the Fazekas scoring system [6]. Volumetric segmentation of periventricular WMH within the ventral part of the left hemisphere was performed semi-automatically through ITK-SNAP (version 3.8.0; www.itksnap.org) [34]. All MRI scans were evaluated by three blinded experienced raters (GSG, MW, FEdL).
(Immuno-)histochemistry
Tissue blocks were embedded in paraffin and sectioned at 4 µm thickness. These sections were stained with haematoxylin/eosin (HE) following standard histology protocols. (Immuno-)histochemistry was performed on adjacent sections for ionized calcium-binding adapter molecule 1 to detect macrophages and microglia (IBA1; rabbit, ab178846; Abcam, Cambridge, UK; 1:2000, RRID: AB_2636859) and glial fibrillary acidic protein to detect astrocytes (GFAP; rabbit, Z0334; Dako, Santa Clara, CA, USA; 1:600, RRID: AB_10013382).
Sections were first deparaffinized in xylene, rinsed through graded ethanol series and ultimately in demi water. For all immunostainings, sections were processed further using a fully automated immunostainer (Lab Vision Autostainer 360; Thermo Fisher Scientific) and the EnVision FLEX visualization system (K8000, Agilent, RRID: AB_2890017), according to manufacturer’s instructions. Briefly: sections were rinsed in EnVision FLEX Wash Buffer (K800721-2; Agilent, Santa Clara, CA, USA) for 5 min, followed by 5 min in Peroxidase-Blocking Reagent, and a 5-min rinse in EnVision FLEX Wash Buffer. Sections were incubated with the aforementioned primary antibody for 60 min. After incubation, sections were rinsed for 10 min in EnVision FLEX Wash Buffer and incubated for 15 min with EnVision FLEX + rabbit (LINKER) (K800921; Agilent, Santa Clara, CA, USA). After another 10-min rinse in EnVision FLEX Wash Buffer, sections were incubated with EnVision FLEX HRP Solution (Agilent, Santa Clara, CA, USA) for 30 min, then another 10-min rinse in EnVision FLEX Wash Buffer. Sections were incubated with a mixture of EnVision FLEX 3,3'-diaminobenzidine (DAB) + and Substrate Solution (Agilent, Santa Clara, CA, USA) for 10 min and rinsed in tap water for 10 min. GFAP sections were counterstained using hematoxylin before dehydration in ethanol and xylene, and cover slipping.
MRI-pathology co-registration and Regions of interest (ROIs)
Prior to the registration of the MRI images to the HE stained sections, the MRI data of the 22 individuals was loaded into MANGO (Research Imaging Institute, University of Texas health science center, TX, USA; www.ric.uthscsa.edu/mango) [15]. The MRI data (T1-weighted, T2-weighted, T2*-weighted, FLAIR) was compared to the HE reference section to select the corresponding 2D MRI slide. The selection of 2D MRI slides was based on the comparison of different anatomical landmarks (i.e., corpus callosum, caudate nucleus, cortex) across MRI-pathology performed by experienced neuroanatomists (GSG, BG). Then MRI data was registered to the HE reference section based on manual landmark selection using a custom MATLAB script (MATLAB R2020a; MathWorks Inc., Natick, MA, USA). Briefly, MRI data was extracted to previously selected 2D axial slices. At least 10 landmarks were selected on both 2D MRI slide and HE reference section. Next, the MRI 2D images were warped [20] and cropped based on the HE reference.
After MRI-pathology registration, NAWM and WMH were manually segmented based on MRI by different experienced assessors who were blinded to presence of hypertension, sex and other risk factors for white matter lesions. The different regions of interest (ROIs) corresponding to NAWM and WMH were defined for further analysis when agreement was met for at least 2 assessors. MRI and (immuno-)histochemistry data from each individual was subsequently clustered into these ROIs for statistical analysis.
Post-processing of sections
All stained sections were digitized on a Pannoramic 1000 slide scanner (3DHISTECH Ltd, Hungary) employing a 20× magnifying objective. All high-resolution digital images (0.25 µm/pixel) were both visualized and exported to tag image file format (TIFF) (1:4 scale, 8-bit, jpeg with 80% compression) using CaseViewer software (version 2.4; 3DHISTECH Ltd, Hungary). Per specimen, all available stainings were co-registered to corresponding HE through a custom written intensity-based automated multimodal registration MATLAB script (MATLAB R2020a; MathWorks Inc., Natick, MA, USA) [19]. Prior to the co-registration, the images were cropped based on automatic tissue detection. Next, the sections stained for either GFAP or IBA1 were registered to the HE reference section. The result of the registration was visualized via an overlay montage of the HE reference section with IBA1 or GFAP. When the automated registration script failed to accurately register (other) (immuno-)histopathology to HE, these were manually registered based on landmark selection in the target-stained section and HE reference section. 1/22 specimen could not be co-registered to its respective IBA1 staining due to imaging limitations.
Detection and quantification of (perivascular) inflammation
After this multimodal registration, stained sections containing periventricular white matter from all individuals were computationally segmented in ImageJ-MATLAB (version 1.53c, National Institute of Health, Bethesda, MD, United States; MATLAB R2020a; MathWorks Inc., Natick, MA, USA) [9] (we used the color deconvolution tool [25] for GFAP). Analyses were performed on aforementioned MRI-manually segmented ROIs (WMH, NAWM) for each individual. Intensity threshold was used to isolate the target staining from background. To account for varying staining intensities across individuals, the intensity threshold was determined by examining mean intensity values of positive stained cells within ROIs across all individuals for IBA1 and GFAP, separately. Threshold settings based on the overall mean intensity were checked on individual basis. IBA1 intensity was used as a marker for microglial activation [10], since it corresponds to increased IBA1 within microglial cells. Stained area was calculated for IBA1 and GFAP. Similarly, GFAP intensity was used as a marker for astrocytic activation [12]. Frequency of positive stained microglial cells (IBA1) was automatically counted by ImageJ (number per mm2). As GFAP-positive astroglia were often found forming scars in WMH, we did not include astroglia count. In our analysis, we also investigated whether microglial cells (IBA1) showed morphological changes. Based on previous research, it is known that resting microglial cells show a ramified morphology that upon activation changes towards an amoeboid/round shape with shorter ramifications [2]. Therefore, circularity (range 0 to 1) was calculated for IBA1-positive microglial cells; higher circularity values correlate to a rounder/more amoeboid morphology, which is indicative of an activated inflammatory state. Additionally, we measured average cellular length by ImageJ Measure Skeleton Length tool [23] to assess changes in microglial ramifications.
In addition to using GFAP staining to assess neuroinflammation in the brain parenchyma, we used GFAP staining to study perivascular inflammation. Perivascular inflammation was studied on astroglia adjacent to blood vessels within 15 µm. Particularly, perivascular inflammation was evaluated based on the following criteria using the range of perivascular inflammation observed across all individuals: grade 0 (none), grade 1 (mild; < 100 perivascular astroglia per mm2), grade 2 (moderate; 100–300 perivascular astroglia per mm2) and grade 3 (severe; > 300 perivascular astroglia per mm2 or presence of astrogliotic scar surrounding the blood vessels). Additionally, for each vessel presence of close-range perivascular inflammation was assessed. We examined perivascular inflammation within uniformly predefined regions in NAWM and WMH to avoid misclassification. After examination, an average was computed for mild, moderate, severe perivascular- and close-range perivascular inflammation per ROI for all individuals. The number of vessels graded did not differ between groups (p = 0.501) nor regions (p = 0.094).
Statistics
Means and SD were calculated for all continuous variables, as well as frequencies and percentages for categorical variables. When assumptions on normality and homogeneity were not met, we used a natural log transformation. We used multivariate analysis of variance (ANOVA) for group comparisons of age, post-mortem delay, body mass index (BMI) and WMH volume. Relationships between categorical variables were explored using a Chi-square (χ2).
Means for neuroinflammatory markers and perivascular neuroinflammation (individuals with hypertension vs. controls, WMH vs. NAWM) were analyzed using ANOVA, controlled for age, sex and fixation-(immuno-)histochemistry interval, with a Bonferroni correction for multiple testing.
To examine the effect of the severity of WMH burden, we stratified individuals in two groups based on their Fazekas score (mild: Fazekas 0–1; moderate to severe: Fazekas 2–3) [6]. Neuroinflammatory markers and perivascular inflammation across WMH burden groups were analyzed using ANOVA, controlled for age, sex and fixation-(immuno-)histochemistry interval, with a Bonferroni correction for multiple testing.
Results were considered statistically significant when P ≤ 0.05. Statistical analysis of the data was performed using IBM SPSS statistics 25 SPSS (IBM Corporation, Armonk, NY, USA).