A Quantitative Postmortem MRI Design Sensitive to White Matter Hyperintensity Differences and their Relationship with Underlying Pathology

Abstract

White matter hyperintensities (WMHs) associate with both cognitive slowing and motor dysfunction in the neurologically normal elderly. A full understanding of the pathology underlying this clinicoradiologic finding is currently lacking in autopsy-confirmed normal brains. To determine the histopathologic basis of WMH seen on MRI, we studied the relationship between postmortem fluid-attenuated inversion recovery (FLAIR) intensity and neuropathologic markers of white matter lesions (WMLs) that correspond to WMH in cognitively normal aging brains. Samples of periventricular (n = 24), subcortical (n = 26), and normal-appearing white matter (NAWM, n = 31) from 4 clinically and pathologically-confirmed normal cases were examined. FLAIR intensity, vacuolation, and myelin basic protein (MBP) immunoreactivity loss were significantly higher in periventricular WML vs. subcortical WML; both were higher than in NAWM. The subcortical WML and NAWM had significantly less axonal loss, astrocytic burden, microglial density, and oligodendrocyte loss than the periventricular WML. Thus, vacuolation, myelin density and small vessel density contribute to the rarefaction of white matter whereas axonal density, oligodendrocyte density, astroglial burden and microglial density did not. These data suggest that the age-related loss of MBP and a decrease in small vessel density, may contribute to vacuolation of white matter. The vacuolation enables interstitial fluid to accumulate, which contributes to the prolonged T2 relaxation and elevated FLAIR intensity in the white matter.

INTRODUCTION

Hyperintense white matter abnormalities on T2-weighted and fluid-attenuated inversion recovery (FLAIR) magnetic resonance imaging (MRI) scans have been shown to adversely impact attention and motor function in the neurologically normal elderly (1–4). Higher signal intensity on FLAIR imaging reflects prolonged T2 relaxation in tissue due to increased free (or unbound) water. Observations of myelin pallor, axonal loss, and gliosis are the most consistently reported pathologic features of white matter hyperintensities (WMHs) in the setting of white matter pathology found in persons with dementia (5–9). The pathological processes that underlie FLAIR intensity in the white matter of neurologically normal elderly individuals who are free of confounding neurodegenerative disease processes are not completely understood. Detailed evaluation of even small numbers of prospectively studied neurologically normal elderly individuals provides a method for determining the pathologic substrate of age-related WMH.

Areas of hyperintensities can be distinguished by location, either by abutting the ventricle (periventricular) or in the subcortical white matter. Regardless of location, areas lacking a bright FLAIR signal are considered normal-appearing white matter (NAWM). Longitudinal evidence suggests the periventricular WMH appear earlier and increase in volume at a faster rate than subcortical WMH (3, 10). These observations suggest that there may be pathologic differences between periventricular and subcortical WMH.

Postmortem MRI of WMH has previously been investigated using a variety of approaches from agar-embedded slabs to whole cadaver imaging (6, 7, 9, 11–13). We developed a postmortem FLAIR sequence to maximize the sampling accuracy of white matter lesions (WMLs) in postmortem tissue that correspond to WMH. Our study design does not require temperature adjustment, agar-embedding, or slabbing of brain material and, more importantly, enables quantitative evaluation of signal intensity. Whole brain or hemibrain imaging allows for better registration with both antemortem MR images and reformatting for close comparison of the same lesions in postmortem tissue. This comparison alleviates uncertainty over perimortem changes in WMH volume that could affect lesion localization.

From a clinical standpoint, investigations into signal intensity are useful for estimating WMH severity. To determine the relationship between signal intensity and underlying pathology, novel digital pathology methods were developed using whole slide scanning and image analysis with Aperio ImageScope. Pathologic studies of WMH have historically been qualitative in nature (5, 14–16), with more recent studies using labor-intensive semiquantitative eye-piece graticule (9) or 8-bit grayscale conversion for the use of light transmittance quantification (17). Digital pathology, which captures microscopic detail in color, can be analyzed with a broader range of approaches to evaluate various tissue components for pathologic-imaging correlations (18, 19).

In this study, normal brains were collected and imaged in the coronal plane that enabled post-hoc re-orientation to the plane of section that matched the plane of sectioning used in brain dissection. Regions of interests (ROIs) were selected from postmortem FLAIR, with subsequent tissue sampling of matched areas for histologic studies and digital pathology analyses. Our goals were to 1) quantitate intensity differences between normal and diseased white matter on postmortem FLAIR, 2) evaluate the relationship between myelin and axons within the WML, and 3) design an intermediary histologic index to facilitate our understanding of associations between FLAIR signal and pathologic findings.

MATERIALS AND METHODS

Subjects

The Mayo Clinic Study of Aging was the main source of study subjects. The postmortem imaging of 41 brains was done in collaboration with the Division of Anatomic Pathology at the Mayo Clinic (Rochester, MN). MRI and pathologic assessment were included in this internal review board-approved prospective study of aging. The brains underwent a postmortem FLAIR protocol on a 3T (GE Signa). Cases were excluded if they had a clinical diagnosis of anything other than cognitively normal and clinical dementia rating scale (CDR) = 0 (n = 25); exclusions included mild cognitive impairment with an amnestic syndrome, probable Alzheimer disease (AD), Lewy body dementia, frontotemporal dementia, dementia-hard to classify, and uncertain. Of the remaining 16 cases, 1 clinically diagnosed normal was excluded for a CDR = 0.5. The length of time between cognitive evaluation and autopsy date was required to be less than 12 months. The time restriction was designed to provide a more accurate depiction of the postmortem pathology found in the white matter of a neurologically normal cohort. Of the 15 normal subjects, 5 remained with a clinical evaluation within 1 year of death. Of the final 5 cases, 1 had to be excluded after subsequent neuropathologic investigation revealed extensive AD type pathology. The 4 cases included in this report had at most mild age-related pathology and no findings sufficient to meet criteria for any other neurodegenerative disorder. Further demographics and postmortem characteristics are described in the Table. Tissue was sampled from areas corresponding to periventricular (n = 24) and subcortical (n = 26) WMH and NAWM (n=31).

Table.

Subject Demographic and Postmortem Characteristics of Neurologically Normal Cases

	Demographics				Postmortem characteristics
Case no.	Sex	MMSE (mos elapsed)	Age at death (y)	APOE	Cause of death	Time from death to pmMRI	Brain weight (g)	Braak NFT stage	CERAD	AGD	LBD
1	F	n/a* (3)	89	n/a	Acute renal failure	7 d	1284	III	Normal	N	N
2	F	28 (8)	87	3/3	Congestive heart failure	7 d	1216	III – IV	Normal	Y	N
3	M	30 (11)	90	3/4	Ischemic heart disease	7 d	1330	III – IV	Normal	Y	N
4	M	28 (12)	74	3/3	Ischemic heart disease	7 d	1268	II – III	Normal	Y	N

Abbreviations: F – Female, M – Male, MMSE – Mini-Mental State Examination (*n/a – neuropsychological testing was not available, perimortem clinical information was evaluated and consistent with neurologically normal cognition), MMSE (months elapsed): time between date of death and clinical evaluation, pmMRI – postmortem MRI, NFT – neurofibrillary tangle, CERAD – Consortium to Establish a Registry for Alzheimer’s disease, AGD – Argyrophilic grain disease, N – No, Y – Yes, LBD – Lewy body disease.

Neuroimaging and Matching Procedure

After remaining in fixative (15% formalin) for 7 days after autopsy, the left hemisphere was rinsed in distilled water and placed in a plastic rectangular container of distilled water. The imager ran a gloved hand along the surface of the brain to displace any air bubbles that might interfere with imaging. Coronal 2D volume postmortem FLAIR was acquired perpendicular to the anterior-posterior axis of the brain in the axial scan plane using superior/inferior (S/I) input for frequency direction. The inversion time (TI) required to null the signal of distilled water (T1 = 3175 ms) at 3T (20) was calculated using the following equation (21):

$$\text{TI}(\text{null})={({\text{log}}_{\mathrm{e}}2)}^{*}\mathrm{T}1$$

$$\text{TI}(\text{null})=(\text{ln}2)(3175\text{ms})=2200\text{ms}$$

In order to achieve a null signal from the distilled water with maximal value on the T2 decay curve, we optimized TR = 15002 ms. To maximize the contrast between the WMH and the white matter, minimum TE was chosen (typically 80 ms). There were 94 partitions, each 1.0 mm in thickness with 1.0 mm gap. The field of view was 18 cm × 18 cm and we used 2 signal averages with a 192 × 192 matrix and a bandwidth at 31.25 kHz.

Five slices were selected from each of the 4 postmortem FLAIRs at the level of the nucleus accumbens, lentiform nucleus, lateral geniculate nucleus, splenium, and primary visual cortex. The software program Analyze (Biomedical Imaging Resource, Rochester, MN) was used to match FLAIR slices and autopsy photographs on a dual-monitor system. The realigned postmortem FLAIR was then used to delineate periventricular, subcortical, and NAWM ROIs. An example of a manually matched slice is illustrated in Figure 1, using the program Photoshop (Adobe, San Jose, CA) to overlay the postmortem tissue with a semi-transparent postmortem FLAIR. Cerebral landmarks (e.g. gyri or deep gray matter) were used to facilitate white matter ROI placement (Figures, Supplemental Digital Content 1, http://links.lww.com/NEN/A411 and Supplemental Digital Content 2, http://links.lww.com/NEN/A412). A periventricular WMH was defined as abutting the lateral ventricular surface. WMH that were not in direct contact with the ventricle or a part of a homogenous continuum extending from the ventricular surface were considered subcortical. NAWM was identified on the postmortem FLAIR as having no increased signal and the border could not be in direct proximity to a hyperintense region of white matter. Nearly every slice demonstrated WMH along the periventricular surface, with the exception found at the parietal lobe. Therefore, to avoid bias based on underlying anatomical differences, NAWM was selected in proximal to both periventricular and subcortical WMH. The white matter ROIs determined on the postmortem FLAIR had to show no evidence of a demyelinating, infectious, toxic, or metabolic process in the postmortem tissue. The Analyze program was used to determine the mean intensity value and area for the periventricular, subcortical, and NAWM ROI on the postmortem FLAIR. In addition, an ROI was drawn to quantify ventricular fluid signal closest to the white matter regions being evaluated. The signal intensity was used to normalize the periventricular, subcortical, and NAWM ROI by subtracting the value, such that:

$${\mathit{\text{ROI Intensity}}}_{\mathit{\text{normalized}}}=\mathit{\text{ROI Intensity}}-\mathit{\text{Ventricular Fluid Intensity}}$$

Figure 1.

A–D The autopsy photograph (A) served as the base to which the postmortem FLAIR (B) was reoriented to match (C). The overlay demonstrates the hyperintense meninges had the same pattern as the postmortem tissue. The periventricular (green arrow) and subcortical white matter hyperintensity (maroon arrow) were seen in relation to the tissue. (D) The dotted lines indicate the region sampled for histologic evaluation.

The ventricular fluid acted as an internal standard for each slice, whether the ROI was in the frontal or temporal lobe, etc.

WML Area and Burden Analysis

Tissue was sampled according to the location of a periventricular, subcortical, and NAWM ROI found across the 5 postmortem FLAIR slices. Five-µm-thick sections were stained with hematoxylin and eosin (H&E), Luxol fast blue-periodic acid Schiff (LFB-PAS), and by immunohistochemistry for myelin basic protein (MBP), phosphorylated neurofilament (pNF), glial fibrillary acidic protein (GFAP), microglia [ionized calcium-binding adaptor molecule (IBA1)], and collagen-IV. Antibody dilutions and manufacturer details are listed in Supplemental Methodsm Supplemental Digital Content 3, http://links.lww.com/NEN/A413.

All slides were digitally scanned at 20x using the ScanScope XT (Aperio, Vista, CA.). ImageScope software (Aperio) was used to draw each ROI on the LFB-PAS slide that corresponded to the FLAIR ROI blind to the intensity level of the corresponding WMH. An area output was extracted from the original ROI and used to compare to the area of the corresponding postmortem FLAIR ROI. Each ROI from the LFB-PAS slide was copied and pasted to the same histologic position across all serially stained sections. Myelin, axonal and astroglial burden were analyzed using the Positive Pixel Count (PPC) algorithm designed to detect the brown hue and antibody saturation of 3,3’ diaminobenzidine (DAB) corresponding to the MBP, pNF, and GFAP, respectively (see Results for optimization of input parameters). The percentage positivities for myelin and axonal staining were normalized to the corpus callosum to control for any inter-subject staining variability. The resulting value was a percentage of positively stained pixels per area annotated. Vacuolation was also quantified using the PPC algorithm, but was modified to determine the percentage of space not occupied by neuropil, cells or vessels.

An advanced nuclear algorithm was modified to design 3 unique algorithms to count small vessels, activated microglia, and oligodendrocyte nuclei (Nuclear Algorithm 2004, Aperio). The algorithm used pixel count, roundness, elongation, and compactness to help detect the object of interest. The output contains a count, as well as area that were used to determine object density to remove ROI size as a confounding variable. Microglia were immunostained with IBA1 antibody (22) and operationally defined using morphologic characteristics and the optical density of the IBA1 stain (Nuclear v9, Aperio).

Statistical Analysis

Sigma Plot 11.0 (Systat, Chicago, IL) was used for all statistical analyses. Associations between antemortem and postmortem FLAIR area and intensity were assessed using Spearman rank order correlations. The non-parametric Kruskal-Wallis test was used to examine significant differences amongst the median values of WML and NAWM for each variable that failed normality and equal variance. Subsequent analyses used the non-parametric Dunn’s method for pair-wise comparisons. ANOVA was performed to assess differences in mean values for oligodendrocyte and small vessel density, which passed both the normality and equal variance tests. The Holm-Sidak method was run for pair-wise comparisons procedures in these 2 data sets. Two multiple linear regression analyses were run to evaluate which pathologic variables were the significant determinants of signal intensity and vacuolation. Both models adjusted for age, sex and tissue archival time (See Supplemental Methods, http://links.lww.com/NEN/A413).

RESULTS

Pathologic Observations

Gross inspection of the white matter at the angle of the ventricle revealed softening with a visibly discrete boundary coinciding to an area of hyperintense signal on the postmortem FLAIR in the frontal, temporal and occipital lobes, but rarely within the parietal lobes. Dilated perivascular spaces were often grossly apparent in the periventricular WML (i.e. periventricular cap). Gross appearances were mostly unremarkable in the subcortical region corresponding to the low signal intensity of diffuse WMH, but mild discoloration and dilated perivascular spaces were apparent in more hyperintense subcortical regions.

Microscopic evaluation of the white matter in areas corresponding to high signal intensity on FLAIR demonstrated rarefaction associated with a loss of MBP immunoreactivity when compared to NAWM (Fig. 2B). Periventricular WMLs were more severely affected than subcortical WMLs, which was consistent with the greater FLAIR signal intensity in periventricular WMH. The LFB-PAS stain of the periventricular WML showed the same degree of myelin loss, but was not as effective in revealing the more subtle myelin pathology in the subcortical WML (Fig. 2A). The axonal density, as detected with pNF immunohistochemistry, was reduced in both the periventricular and subcortical WML and to a greater extent in the periventricular areas of hyperintensity. An overall decrease in pNF was observed in both periventricular and subcortical WML (Fig. 2C), coinciding with MBP loss. Vacuolation of neuropil was prominent in the periventricular WML and exaggerated in the perinuclear region of oligodendrocytes (Fig. 2D).

Figure 2.

A, B A graded decline in myelin from normal white matter to white matter lesions was visualized at 20x magnification using the special stain Luxol fast blue-periodic acid Schiff (LFB-PAS) (A) and immunohistochemistry for myelin basic protein (MBP) (B). (C, D) Axon loss was most pronounced in the periventricular area, suggesting that vacuolation of the white matter may attributable to myelin loss. pNF, phosphorylated neurofilament; H&E, hematoxylin and eosin; NAWM, normal-appearing white matter. Scale bar = 50 µm.

The periventricular white matter most proximal to the ventricle has been described as a highly cellular region separated from the ependymal lining by a gap of GFAP-positive fiber bundles (23). Astrocytic gliosis extended beyond this region and throughout the periventricular WML (Fig. 3A), but was not confined to the periventricular area of myelin and axonal loss. The reactive astrocytic processes lessened considerably in the subcortical white matter and were not observed to associate with the subcortical signal intensity on postmortem MRI. Perivascular astrocytic gliosis was noted to associate with abundant corpora amylacea.

Figure 3.

A, B Periventricular white matter lesions (WML) had the highest astroglial burden, microglial (Iba-1-immunopositive) density, and oligodendrocyte density but there were no significant differences between the subcortical WML and the normal-appearing white matter (NAWM). (C) There are fewer small vessels in the WML compared to the NAWM. GFAP = glial fibrillary acidic protein. Scale bar = 50 µm.

Amoeboid-like microglial cell bodies with retracted and thickened cell processes were found dispersed throughout the neuropil of periventricular WML, but had a lower burden in the subcortical WML and NAWM (Fig. 3B). The processes were also thicker in the periventricular WML compared to the subcortical WML and NAWM. The microglia did not show preferential localization to perivascular spaces. The oligodendrocyte population was qualitatively less in the periventricular WML compared to both the subcortical WML and NAWM, but was difficult to discern upon inspection of hematoxylin-counterstained IBA1 slides. Collagen-IV-positive small vessels were not greatly different between WML, but appeared higher in the NAWM (Fig. 3C).

Quantitative MRI

Area measurements of WMH on the postmortem FLAIR were significantly correlated to the corresponding WML area measurements on the histologic section. Periventricular WMH had a more clearly demarcated boundary compared to the diffuse nature of subcortical WMH (Fig. 1). Higher periventricular area measurements on coronal postmortem FLAIR slices correlated significantly with higher areas measured on matched histologic sections (r = 0.83, p < 0.001). Despite the lack of clearly defined borders, a higher subcortical area on the postmortem FLAIR was significantly associated with matched subcortical sections in its corresponding histologic section (r = 0.72, p < 0.001). The mean intensity signal measured in the periventricular and subcortical WMH did not correlate with either postmortem area on FLAIR or in the histologic white matter lesion. The intensity differences observable in Figure 1 were significantly different between normal and diseased white matter (p < 0.001, Fig. 4A). The periventricular WMH had the highest signal intensity and NAWM, the lowest. Though a gradual difference between tissue types has been suggested (17), the overlap observed in the box plots could be explained by the highly inter-related nature of the WMH types (p < 0.0001). Thus, the higher NAWM values can be seen in the cases with higher subcortical and periventricular WMH.

Figure 4.

Median values were compared to determine quantitative differences between periventricular and subcortical white matter lesions and normal-appearing white matter (NAWM). (A–H) Postmortem MRI intensity (A), vacuolation (B), myelin burden (C), axonal burden (D), oligodendrocyte density (E), astroglial burden (F), microglial density (G), and small vessel density (H) were analyzed using a Dunn’s method pairwise comparison. Note the pattern of significance among the pathologic variables. †, Periventricular values were significantly different than both subcortical and NAWM; ‡, subcortical values were significantly different than both periventricular and NAWM; ∫, NAWM values were significantly different than both periventricular and subcortical white matter.

Digital Microscopy Modifications

An intermediary histologic index was designed to facilitate our understanding of associations between FLAIR signal and pathologic findings. To quantify the percent area unoccupied by neuropil, cells or vessels, a PPC algorithm was manipulated (Figure, Supplemental Digital Content 4, http://links.lww.com/NEN/A414). All H&E slides were stained using an automated machine, but to avoid the influence of tissue staining differences the entire spectrum of colors (e.g. 360° color wheel) had to be analyzed and not just the pink/red hues of an H&E stain. The algorithm was still able to detect intensity level differences [i.e. 0 (black) − 255 (white)] in the stained tissue. We found that we were able to discern unoccupied space accurately by maximizing the upper limit of the weak positive pixels, setting the lower limit of the weak and medium positive pixels equal to each other, and minimizing the lower limit of the strong positive pixels. Finally, we found that the algorithm had to be manipulated to ignore what would normally be considered “negative pixels” by minimizing the color saturation threshold. The resultant vacuolation was calculated by:

$$[(\#\text{weak positive})/(\#\text{strong positive pixels})]\times 100=\%\text{vacuolation}$$

Collagen-IV-immunopositive vessels were visible in both cross-sectional and longitudinal view. We found that relative to the default parameters of the nuclear algorithm: a higher radius (3 µm) for noise reduction was necessary, a nearly maximum curvature threshold prevented large vessels from being counted, a minimum of 350 pixels, and a higher compactness to avoid inaccurate labeling of background enabled quantification small vessels. Oligodendrocytes have prominent rounded nuclei with tightly compacted basophilic chromatin helping to distinguish them from astrocytes, which tend to be larger, pale and less compact (24, 25). This required a high roundness, compactness and elongation values sensitive to the hematoxylin (blue) counterstained nuclei. The oligodendrocyte algorithm was run on GFAP, IBA1 and collagen-IV immunostains to ensure specificity in lieu of an oligodendrocyte marker. The oligodendrocyte algorithm did not label astrocytes, microglia, or vessels on any the respective stains. The microglia algorithm was designed to distinguish the cells by their larger amoeboid-like cell body with retracted and thickened cell processes. Relative to the default parameters, experimental modifications that enabled labeling of the amoeboid-like microglia included a higher averaging radius (5 µm), a higher curvature threshold to exclude “resting” microglia with extended processes, optical density values specific to IBA1-immunopositive microglia, and maximum threshold values using the edge threshold method. Density measures were calculated using:

$$[(\text{Number of objects})/{\text{Area in mm}}^2)]=\#\text{Objects}/{\text{mm}}^2$$

FLAIR Intensity-Pathology Relationships

To assess pathologic changes that underlie signal intensity, group differences by WMH type were analyzed using quantitative pathologic data. Comparisons are illustrated as box-plots in Figure 4. As expected, measures of vacuolation and myelin were higher in the periventricular WML and lowest in the NAWM (p < 0.001) (Fig. 4B, C). Axonal loss, oligodendrocyte loss, astrocytic burden and microglial density were greatest in the periventricular WML compared to both the subcortical WML and NAWM (p < 0.05), which were not found to significantly differ (Fig. 4D–G). Small vessel density was significantly lower in the WML compared to the NAWM (p < 0.05) (Fig. 4H).

A Spearman correlation run across all WMH types revealed that each quantitative pathology measure correlated with signal intensity (p < 0.007), except small vessel density. To evaluate which pathologic measure was the strongest determinant of signal intensity, a linear regression model was constructed adjusting for age, sex and tissue archival time. The FLAIR signal intensity (dependent variable) could be predicted from a linear combination of myelin burden (p < 0.001) and small vessel density (p = 0.015), but none of the other measures. Vacuolation was found to associate with FLAIR signal, but is not itself considered a pathologic entity and therefore not included in the model. Instead, to test the hypothesis that FLAIR intensity acts as a surrogate for underlying tissue vacuolation, a similar model was constructed with vacuolation as the dependent variable. Supporting our hypothesis, vacuolation was also found to be predicted by myelin burden (p <0.001) and small vessel density (p = 0.015), with no contribution from the other measures.

DISCUSSION

We characterized and quantified neuropathologic changes associated with WMH in a neurologically normal cohort who had no significant neurodegenerative pathologic changes. Our main findings were that vacuolation, myelin density and small vessel density contribute to the rarefaction of white matter in neurologically normal elderly individuals. Axonal density, oligodendrocyte density, astroglial burden and microglial density did not. Oligodendrocytes were not significantly decreased in the weakly hyperintense subcortical WML, suggesting that myelin proteins are adversely affected first with subsequent involvement of oligodendrocytes. Astrogliosis and activated microglia were restricted to the periventricular region, suggesting their role may be more regional and not fundamental to MRI WML.

Although comparisons between stained sections and FLAIR demonstrates postmortem MRI may not be sensitive to milder white matter pathology (7), our goal was to offer evidence for pathologic interpretation of age-associated WMH. Other studies have used postmortem MRI procedures to localize WML (6, 7, 9, 15, 26), but their mixed cohorts make interpretation of results hard to apply to a neurologically normal elderly patient population. AD is known to affect axonal density and heighten gliosis in the white matter (11, 17, 27), potentially confounding age-related pathologic interpretation of FLAIR signal changes in periventricular and subcortical WMH relative to NAWM. Evaluation of periventricular WMH is facilitated by using the lateral ventricle border as a consistent landmark, but our results support evidence that postmortem FLAIR imaging is a useful tool for locating both periventricular and the more elusive subcortical white matter lesions in postmortem tissue (6, 9, 12, 17). In order to evaluate histologic-imaging relationships, we investigated intensity differences between periventricular and subcortical WMH in all 4 lobes of the brain. We reported a robust correlation between imaging and histologic area measurements. Our findings supported the notion that radiologic (i.e. intensity) differences exist between periventricular and subcortical WMH (28), which were shown to associate with myelin loss, small vessel loss and increased vacuolation in an elderly cohort of cognitively normal individuals (15). We evaluated severity of myelin and axonal loss in periventricular and subcortical lesions and show axons in the subcortical WML remain fairly resistant to the age-related pathology affecting myelin. Axonal loss has been inconsistently reported, but may be related to mixture of dementia and control cases included in previous studies (5, 11, 17).

Histologic differences between periventricular and subcortical WML may result from temporal duration of pathology, spatial localization within the white matter, or a combination of both (28). Temporal duration leading to histologic differences is supported by longitudinal evidence of periventricular WMH appearing before subcortical WMH and increasing at a faster rate (10). Our results support the hypothesis that higher signal intensity reflects higher white matter vacuolation, which is a byproduct of myelin and small vessel loss. The obliteration of the small vessels has been shown in aging brains (29), using 3-dimensional celloidin sections highlighting what are termed “string vessels”. Small vessel density measures were much lower in both periventricular and subcortical WML compared to NAWM, supporting a vascular contribution to white matter pathology.

Despite the loss of myelin in subcortical WML, the oligodendrocyte population remained unaffected compared to concomitant loss of both myelin and oligodendrocytes in periventricular WML. This might suggest that myelin is more susceptible to mild pathologic insult in subcortical WML, perhaps via excitotoxic injury. NMDA receptors are preferentially located in the distal oligodendroglial processes that form myelin (30, 31). Significantly more activated microglia were found in the periventricular WML compared to the subcortical WML, supporting previously reported associations (32). Greater microgliosis may contribute to oligodendrocyte loss in periventricular WML, via release of proinflammatory cytokines or other substances (33).

Astrogliosis appeared to follow the tract of the association fibers, but was not considered to contribute to the hyperintense nature of periventricular WML as it was shown to extend past areas of myelin loss. Previous studies have not reported a difference in astrocytic burden between WML and NAWM, but this may be related to inclusion of cases with concomitant neurodegenerative pathology (5, 17, 26). It has been suggested that periventricular astrogliosis is a function of ependymal disruption (11, 14, 34); however, an early diffusion study showed the CSF-brain interface is permeable to substances entering/exiting the CSF (35). The apparent ease of diffusion suggests “leakage” of the CSF is not likely the culprit eliciting an astroglial response, when the ependymal lining was not designed to be a structural barrier. Supportive evidence for a disassociation between denudation of the ependymal lining and astrogliosis comes from the periventricular parietal white matter region at the level of the splenium. These regions were often observed to have no hyperintense signal on FLAIR, but a high astroglial burden even in areas where microscopic inspection revealed the discontinuity of ependymal cells. While GFAP is the stain of choice for fibrillary astrocytes in the white matter (24, 36), we acknowledge a basal level of GFAP-immunoreactivity is expected in white matter. There were qualitatively less GFAP-positive astrocytic cell bodies in these parietal periventricular white matter regions, but still showed a large number with GFAP-positive astrocytic processes.

While we recognize the limitation of our sample size, this study provides quantitative neuropathologic and radiologic measures across multiple regions. The study sample was derived from a consecutive series of autopsied cases. We applied stringent inclusion requirements to assess the pathology underlying WMH in a cognitively normal aging brain. In order to overcome this limitation we sampled multiple regions, bringing our dataset to a total 24 periventricular and 26 subcortical ROIs. Another potential limitation of this study is the quantitative analysis of FLAIR signal intensity. To ameliorate differences across cases, each brain remained in fixative for the same length of time prior to imaging (7 days), was handled the same way in preparation of imaging in distilled water, and was normalized to the ventricular fluid of each slice in proximity to the white matter regions being evaluated. The WML examined in the postmortem tissue were not discrete lesions consistent with a demyelinating, infectious, toxic or metabolic process. There was no apparent swelling of the tissue, which would be reflective of an edematous process that could result in a dilution effect of the tissue components in the expanded interstitial space. Even with auto-staining techniques, staining differences can occur as a result of a number of variables such as length of time in fixation. In order to overcome these differences, the algorithms were designed to consider a range of staining intensity as positive and only ignore pixels outside of the specific hue, for example, brown chromogen, DAB.

The signal intensity was observed to be a direct reflection of myelin and small vessel density loss that acts as a surrogate for vacuolation. This would suggest a higher signal would be interpreted as a greater loss of myelin and small vessels, which should be considered in volumetric MRI studies quantifying WMH volume with relation to clinical phenotypes. When assessing WMH volume there is typically an assumption that all voxels of WMH represent the same degree of underlying pathology, which may confound results. One group included a qualitative categorization based upon signal intensity (37), but to our knowledge the field lacks a study that examines the impact signal intensity may have on functional clinical measures. We propose an index of signal intensity be evaluated in relation to clinical impact as a measure of pathologic severity in both an antemortem series and a larger series of autopsied cases with postmortem FLAIR. Additionally, future studies should address what category of small vessels is lost or if it is a uniform involvement of arterioles, venules and capillaries.