Post-mortem correlates of Virchow-Robin spaces detected on in vivo MRI

The purpose of our study is to quantify the extent to which Virchow-Robin spaces (VRS) detected on in vivo MRI are reproducible by post-mortem MRI. Double Echo Steady State 3T MRIs were acquired post-mortem in 49 double- and 32 single-hemispheric formalin-fixed brain sections from 12 patients, who underwent conventional diagnostic 1.5 or 3T MRI in median 22 days prior to death (25% to 75%: 12 to 134 days). The overlap of in vivo and post-mortem VRS segmentations was determined accounting for potential confounding factors. The reproducibility of VRS found on in vivo MRI by post-mortem MRI, in the supratentorial white matter was in median 80% (25% to 75%: 60 to 100). A lower reproducibility was present in the basal ganglia, with a median of 47% (25% to 75%: 30 to 50). VRS segmentations were histologically confirmed in one double hemispheric section. Overall, the majority of VRS found on in vivo MRI was stable throughout death and formalin fixation, emphasizing the translational potential of post-mortem VRS studies.


Introduction
Virchow-Robin spaces (VRS) are the anatomical spaces between the basement membrane of the perivascular astrocytic feet processes (glia limitans perivascularis) and the basement membranes of the blood vessel endothelia. 1 Within the VRS, CSF flows from the subarachnoid space into the brain parenchyma, allowing for exchange with brain interstitial fluid (ISF). Additionally, VRS are involved in the intramural, perivascular connective tissue drainage pathway, from the brain, towards the subarachnoid space. 2 They are thus involved in central nervous system (CNS) debris removal and lymphatic immune gateways. 3 In healthy control subjects, the size of VRS is below the resolution of current conventional brain MRI scans at 3T. 4 However, large Virchow-Robin spaces (VRS) are a frequent imaging finding in routine clinical MRI scans of the brain. 5 Enlargement of the perivascular spaces has been suggested as a consequence of debris accumulation, 6 e.g., in the course of mucopolysaccharidoses 7 or during collagen deposition in atherosclerosis, 8 but no singular mechanism has emerged as yet. In multiple sclerosis, a disease with chronic inflammatory and neurodegenerative components, 9 inflammation, 10 in line with community-based studies of ageing, 11 as well as dilatation associated with chronic brain parenchymal atrophy, 12 were associated with enlarged VRS 13 and perivascular collagen deposition. 14 Large VRS are associated with an increased risk for lacunar stroke, 15 stroke recurrence, 16 vascular-related cognitive impairment, 17 oral anticoagulation-related symptomatic cerebral hemorrhage, 18 and progression of disability in Parkinson's disease, 19 as well as both present and future cognitive impairment. [20][21][22][23] Detailed histological and molecular analysis of large VRS on post-mortem tissue could further guide our understanding of CNS waste clearance and the influence of vascular pathology on CNS disorders. However, the actual histopathological correlates of VRS seen on in vivo MRI has remained largely unspecified, as it is still unclear at which rate large VRS were preserved after death and post-mortem tissue processing.
The purpose of our study therefore was to quantify the extent to which VRS detected on in vivo MRI are reproducible by post-mortem MRI in 12 individuals who underwent MRI at a median of 22 days before their death and in whom the brain was available for post-mortem processing.

Study cohort
Individuals were recruited based on the following criteria: brain MRI within <one year before death; in vivo MRI with sufficient image quality to visualize and segment the VRS; and availability of post-mortem brain tissue from the same patient at our local neuropathology department. This inclusion criterion was met by twelve consecutive patients. Clinical details, including the number of analysed tissue blocks per patient, main clinical and neuropathological diagnoses, the cause of death, the time intervals from in vivo MRI to death, and from death to post-mortem MRI, are summarized in Table 1.
The tissue slicing was performed in the course of neuropathological clinical routine at standardized section levels, including coronal sections at the frontal poles, at the level of the temporal poles, a mid-thalamic section level and occipital sectioning. The availability of these tissue sections for this study was determined by prior neuropathological diagnostic work-up, regions of clinical interest, typically those including lesions, were utilized and not available for our subsequent analysis. The entire available brain tissue was analyzed in each patient. In total we investigated 81 tissue-blocks, of which 49 were double-and 32 single-hemispheric/ lobar formalin-fixed brain sections; On a patient basis in median 4 double-hemispheric sections (25%-75%: [3][4][5], and 2 single-hemispheric sections were included (25%-75%: 0-5) Table 1.

MRI acquisition in vivo
In vivo MRI acquisition was performed in the course of the clinical routine, with imaging adapted to the clinical question and thus not standardized across all patients. However, the imaging protocols were quite similar, as all patients were imaged in our department. A 1.5 Tesla Philips Ingenia MRI with an eight-channel head coil was used in 10/12 subjects. The remaining two subjects were scanned with a Philips Achieva (3T, 32-channel head coil) and a Siemens Vida (3T, 64-channel head coil), respectively. 3D-T1-weighted images were available in all cases. A detailed list of the analyzed in vivo MRI sequences is available in Supplementary Table 1.

MRI acquisition post-mortem
From the included 12 subjects, a total of 49 doubleand 32 single-hemispheric/lobar formalin-fixed brain sections, with a slice thickness of around 1 cm were obtained for the post-mortem MRI analysis, the sections were cut prior to formalin fixation. Tissue sections were axially oriented in one subject (#22-19), which was anaylsed axially, wherease coronal reformates were used in all the other cases. Post-mortem scanning was performed with a Siemens Vida 3T scanner.
Formalin-fixed tissue blocks were individually placed in a flat plastic container and fully emerged in Flourinert. 24 Double Echo Steady State (DESS) images were acquired using the built-in spine coil of the table, which has 72 channels in 12 segments, of which we used four, and the Siemens "multiflex surface coil," with 18 channels on top of the specimen.
T2-weighted DESS images were acquired with an echo time of 5.6 ms, a repetition time of 17 ms, and a flip angle to 25 . The isotropic resolution was 0.4 mm (matrix: 468 Â 576), and the acquisition time, per tissue block, was around 30 to 50 min. Due to the immersion in Flourinert the brain is surrounded by extremely low signal intensity despite the T2-weighted contrast.

Histology acquisition
To test if VRS segmented on MRI correspond to VRS histologically, proof of concept was tested in one double hemispheric section from ID #120-19. The section was stained with Hematoxylin Eosin, Elastica van Gieson, and Luxol fast-blue myelin stain as reported previously. 25 Histological images were manually co-registered to the aligned post-mortem and in-vivo MRI images.

Image analysis
In a first step, VRS were manually segmented with 3D-Slicer 26 on the in vivo data set (AE, SM, LH) and subsequently on the post-mortem set (AE, SM, LH) in the entire brain. Lesions were masked from the analysis (LH). The in vivo segmentation were performed without knowledge of the post-mortem images. As suggested previously (STRIVE-criteria), VRS were defined as CSF iso-intense lines following the course of vessels in the supratentorial white matter and deep gray matter. 1 The post-mortem MRIs were manually and rigidly co-registered with the corresponding in vivo MRIs based on anatomical landmarks with Freeview/ FreeSurfer. 27 VRS segmentations were not used as additional information/landmarks for the registration process.
The amount of overlap of the in vivo and the postmortem VRS segmentations were manually quantified for each tissue-block using Freeview/FreeSurfer by LH/ SM. To account for the inherent tissue distortion that occurs between in vivo and post-mortem imaging and the limited slice thickness ($1cm), the analysis was limited to one section level within each tissue-block with the highest spatial correspondence of anatomical landmarks. VRS were counted on in vivo MRI scans with both segmentations from in vivo and post-mortem MRI overlaid at 50% opacity in Freeview. We collected the counts of VRS where the segmentations overlapped or did not overlap for each tissue block ( Figure 1).

Statistical analysis
The statistical analysis was calculated with R-studio. 28 The ratio of in vivo VRS with and without postmortem correlates was collected on a tissue-block basis and averaged per patient for two anatomical areas: deep gray matter (including thalamus or putamen and globus pallidum) and supratentorial white matter. The rates are provided with medians and a 25% to 75% range.
In tissue blocks with no VRS on either in vivo or post-mortem MRI, which was the case in 4/20 blocks with deep gray matter, we did not calculate or assign a detection rate.
To account for the potential effect of confounders on the post-mortem detection rate of VRS as imaged on in vivo MRI, we used linear mixed-effect regression models. These models were separately calculated for supratentorial white matter and deep gray matter. Fixed-effect variables included in these two models were the in vivo MRI-to-death interval [days], the death-to-post-mortem MRI interval [days], whether tissue-blocks were single-hemispheric/lobar or doublehemispheric, and the surplus of VRS found on postmortem MRI relative to the total number of VRS in this tissue block, ie., [(post-mortem -in vivo)/(postmortem þ in vivo)] and the total number of tissue blocks analyzed per patient . These models used the individual tissue block as the unit of analysis. To account for the nested structure of the data, tissue blocks were nested within patients. We report the beta for each predictor with its 95% confidence interval and p-value.
Normality was visually assessed with quantilequantile-and density plots and qualitatively with the Shapiro-Wilk test.
Inter-rater agreement was assessed with inter-class correlation coefficients, which are provided with their respective 95% confidence interval and p-value.

Data availability statement
Anonymized data, not published in the article, will be shared upon reasonable request from a qualified investigator after approval from the local Ethics committee.
The study was approved by the ethics committee of the Medical University Vienna and performed in accordance with the Helsinki Declaration. Written, informed consent was obtained from all subjects or their legal representatives as far as required by Austrian law.

Cohort characteristics
Twelve subjects contributed a total of 49 double-and 32 single-hemispheric/lobar formalin-fixed brain sections. The median age at death was 62 years (25% to 75%: 50 to 77-years). The time from MRI to death was a median of 22 days (25% to 75%: 12 to 134 days). The formalin fixation time at post-mortem MRI acquisition was a median of 236 days (25% to 75%: 136 to 407 days). Subjects (female: male ¼ 3: 9) died during the course of various conditions, including malignant/metastatic disease (n ¼ 8), decompensated hepatic cirrhosis with cardio-pulmonary resuscitation (n ¼ 1), cerebral vascular occlusion (n ¼ 1), constrictive pericarditis and cerebral tuberculosis (n ¼ 1), and amyloid angiopathy (n ¼ 1) (see Table 1 for details). Supratentorial white matter was present in all tissue blocks (n ¼ 81) and thus in all subjects, the deep gray matter was present in 20/81 tissue blocks and absent in three subjects.
Detection rates of in vivo MRI VRS by post-mortem MRI. Supratentorial white matter was present in all tissue blocks (n ¼ 81) and thus in all subjects.
On a patient basis, a median of 80% (25% to 75%: 60 to 100) of VRS found on in vivo MRI were reproduced at post-mortem imaging in the supratentorial white matter, which was in absolute numbers a median of 15 VRS/patient (25% to 75%: 9 to 36).
The deep gray matter was present in 20/81 tissue blocks and absent in three subjects.
Overall, the percent of in vivo VRS reproduced on post-mortem images was lower in the deep gray matter than in the white matter, with a median of 47% (25% to 75%: 30 to 50), and a median of 1.5 VRS/patient (25% to 75%: 0 to 5.8) (Figure 2).

Influence of confounding factors on the detection rate of in vivo MRI VRS by post-mortem MRI
To assess the potential influence of confounders on the in vivo -post-mortem VRS consistency, nested linear mixed-effect regression models were computed separately for white matter and deep gray matter ( Table 2).
On an individual block basis (supratentorial white matter n ¼ 81, deep gray matter n ¼ 20), there were no effects of the time-interval from in vivo MRI to death or the time-interval from death to post-mortem MRI on the in vivo -post-mortem VRS consistency.
Similarly, in both locations, the surplus of VRS found on post-mortem images, relative to the total number of VRS in this block, was not related to the in vivo -post-mortem VRS consistency rate.
The total number of tissue-blocks analyzed per patient was not related to the in vivo -post-mortem VRS consistency (deep gray matter model: p < 0.663; white matter model: p < 0.389).

Histological validation
A histological proof-of-concept was achieved in one double hemispheric section from subject #120-19, (Figure 3).
Numerous VRS were perceivable on in vivo MRI and post-mortem MRI (Figure 3) of which one VRS around a perforating artery in the deep white matter of the right superior/medial frontal gyrus was present in the histological section level (Figure 3). Elastica van Gieson staining shows media hyalinosis and extensive perivascular collagen deposition with filling with expansion of the perivascular space. The adventitia is thickened and the surrounding tissue shows slight perivascular tissue rarefaction.
Comparing post-mortem MRI (Figure 4(a), right superior frontal gyrs from subject 120-19) with corresponding histological images (Figure 4(b), luxol fast blue staining) a close spatial correlation between the segmented VRS on MRI and histology is observed (Figure 4

Discussion
Virchow-Robin spaces (VRS) are increasingly recognized to be involved in the pathogenesis of various neurological conditions. While VRS are readily visualized on in vivo MRI, the histo-biological correlates of those VRS seen on in vivo MRI remains uncertain.
Comparability between in vivo imaging and post-mortem analysis is a necessary prerequisite for further investigation of the histological and biological basis of pathologies that originate in or involve this specific brain compartment. In the present cohort we therefore tested the hypothesis if VRS detected on in vivo MRI are reproduced by post-mortem MRI. In 12 subjects with 81 tissue blocks, the majority of VRS detected on in vivo MRI were reproducible at postmortem imaging, providing the first quantification of the translational potential of post-mortem VRS studies.
Large VRS have been associated, among other conditions, with stroke, 15,16 hemorrhage, 18 and cognitive impairment. 17,[19][20][21][22][23]29 Further understanding of the histological and biological processes that occur within large perivascular spaces could thus facilitate our understanding of relevant disease processes. Their enlargement on in vivo MRI has been suggested to occur as a consequence of debris accumulation, 6,29 but VRS are also considered to be fluid-filled spaces 1 and perivascular collagen depositions have been reported in the perivascular space. 8,14 While the histological validation in our study was limited to 1/12 cases, we found that perivascular collagen depositions were associated with VRS that remained visible from in vivo imaging to post-mortem imaging.
Within the supratentorial white matter, where VRS are most abundant, 30 80% of VRS seen on in vivo MRI were also observed on post-mortem images. This is a high detection rate considering the underlying variability (see Limitations section). In the deep gray matter, the rate of the VRS seen on in vivo MRI post-mortem was 47%. The lower reproducibility in the deep gray matter could be due to the smaller sample size of this structure and inherent technical reasons, e.g., the registration technique, manual alignment of anatomical landmarks, could be more efficient with gyri and sulci than with deep gray matter structures, and thus, favor alignment of supratentorial white matter. The lower reproducibility of deep gray matter VRS could also reflect biological differences in the VRS. Histologically, VRS in the deep gray matter are distinct from those in the cortex/white matter, as they are surrounded by two distinct layers of leptomeninges separated by a perivascular space that is continuous with the subarachnoid space. 31 They are also clinically distinct, as large VRS located around the perforating lenticulostriate arteries are frequently observed due to their developmental origin, associated with glial tunnels during axonal growth of the forming anterior commissure. 32 This study has several limitations. Although VRS were rated within the same patient in vivo and postmortem, the interval from in vivo MRI-to-death was short (a median of 22 days), and, although postmortem images were carefully aligned with in vivo MRI (Figures 1 and 3), alignment is never perfect. This has several reasons and major implications for the interpretation of our results. In the present cohort, brain pathology was directly related to morbidity. Thus, even in a median 22-day interval from in vivo MRI to death, some of the in vivo dynamics in VRS remain uncertain. However, we did not observe significant effects of the in vivo MRI to death-and the death to post-mortem MRI time-intervals or other potential confounders, such as the surplus of VRS found on post-mortem images on the overall in vivo -postmortem VRS consistency rate, using nested linear mixed-effect regression models. Furthermore, the brain is distorted in the course of tissue processing and fixation, with around 8% global volume shrinkage over a 70-day fixation period. 33 Further, when fresh brain is cut into slices, the section level is not strictly within the coronal plane that was defined by in vivo MRI. To some extent, this can be corrected with careful manual alignment guided by anatomical landmarks, such as gyri and sulci in 3D data sets, as available in our study. To further account for tissue distortion, the analysis was limited to one section level within the tissue block with the highest anatomical correspondence with in vivo images. Given the post-mortem slice thickness of around 1 cm and the longitudinal course of VRS, we could thus achieve a representative quantification for each block, while preventing overfitting. We further tried to avoid over-fitting by applying a registration that allowed movement within the three spatial directions and adjustment of the overall size of the post-mortem block, but we did not allow local distortion/spatial transformations. Thus, as also visually evident from Figures 1 and 3, we can only provide a rough estimate for the rate at which in vivodefined VRS could be found post-mortem. Another limitation of our study is the sample size (12 cases within a total of 49 double-and 32 singlehemispheric/lobar tissue blocks) with its inherent statistical limitations. In order to maximize the available tissue, we thus included all sections available, leading to differences in the total number of analysed tissue-sections across subjects. In median 4 doublehemispheric sections (25%-75%: 3-5) and 2 singlehemispheric sections were included (25%-75%: 0-5). However, the total number of sections analysed per patient did not show significant effects on the in vivo -post-mortem VRS consistency rates.
The biological processes that lead to the enlargement of VRS will likely differ, e.g., active perivascular inflammation vs. atrophy vs. perivascular collagenosis, etc., and we can not account for this heterogeneity in this cohort.
We observed higher counts of VRS post-mortem than in vivo (Figures 1 and 3), which is not surprising given the higher resolution applied post-mortem (0.4 mm vs. 1 mm isotropic voxel size, 3T vs. mainly 1.5T in vivo) and overall inter-rater agreement was high (>90% both in vivo and post-mortem). However, we can neither exclude artificial "VRS opening" in the course of tissue fixation or ascertain the rate of VRS that may have evolved between in vivo MRI and death. While the scanners were theoretically available to scan the blocks on the same machine on which the in vivo images were acquired, the nature of formalin tissue fixation does not allow acquisition of the identical sequences used in vivo and thus hinder a 1:1 comparison. 34 The proof of concept, that the VRS segmented on MRI correspond to enlarged perivascular spaces histologically was performed in on patient (#120-19), Figures 3 and 4. We observed a close correlation between histologically confirmed VRS and postmortem MRI segmentations in this subject (Figure 4 (c) and (d)). While depositions of collagen, filling and expanding the perivascular space, were observed in this subject (asterisk in Figure 3(f)) that might explain the persistency of these spaces across time and during tissue fixation, we can not extrapolate this observation to other individuals. In line we did not investigate histological factors that influence VRS in vivo vs. postmortem persistency rates and it was unfortunately not possible for us to ascertain if perivascular space enlargement was peri-arterial or peri-venous. Case 120-19 was chosen because death was related to an unexpected sudden event (haemorrhage during thrombectomy), rather than a progressive disease process, and because of the overall high tissue and MRI quality. The high number of VRS on post-mortem scans could arguably also increase the in vivo post-mortem detection rate via "false-positive VRS.", however this theory is not supported by our data, since, on a block basis, the total number of VRS, neither in vivo or post-mortem, were not predictive for the in vivo post-mortem VRS detection rate.

Conclusion
In conclusion, the majority of VRS found on in vivo MRI, was observed on post-mortem imaging, thus suggesting that histological and biochemical analysis of VRS detected by post-mortem MRI could guide future research in this evolving field.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by the Austrian "Multiple Sklerose Forschungsgesellschaft" and LH was supported by the ECTRIMS/MAGNIMS -research fellowship.