Novel Method to Quantify Neuropil Threads in Brains from Elders With or Without Cognitive Impairment

Abstract

Pathological alterations in dendrites and axons (i.e., neuritic pathologies) occur in the normal aging brain as well as in brains from elders with mild cognitive impairment and neurodegenerative dementia. These alterations may correlate with clinical measures of cognitive abilities, but the contribution of neuropil threads (NTs), which constitute 85–90% of cortical tau pathology, has not been clear because of the lack of quantitative methodologies. We combined quantitative fractionation and image analysis to devise a strategy for measuring the burden of tau-rich NTs in the entorhinal and perirhinal cortex of brains from elders with and without cognitive impairment, including dementia due to Alzheimer's disease (AD). On the basis of data presented here using this novel strategy, we conclude that this quantitative imaging technique will facilitate efforts to determine the behavioral correlations of neuritic lesions in AD and other brain disorders.

Symptoms of neurodegenerative dementias begin with an insidious preclinical phase before the emergence of diagnostic clinical features. This phase may be paralleled by the slow accumulation of the characteristic neuropathological lesions of these diseases, which may precede the clinical expression of dementia by years (Morris et al. 1989; Mufson et al. 1999). The most common neurodegenerative dementia is Alzheimer's disease (AD), which is characterized neuropathologically by tau-rich neurofibrillary lesions, extracellular amyloid-β (Aβ) plaques, neuron loss, and gliosis (Braak and Braak 1991, 1994; Hong et al. 2000; Iwatsubo 2000; Masliah and LiCastro 2000).

Although earlier proposals for standard neuropathological criteria for the diagnosis of definite AD relied primarily on Aβ senile plaques (SPs) (Khachaturian, CERAD), subsequent studies have shown that Aβ SPs may be very abundant in the brains of aged individuals who do not exhibit cognitive impairment (Crystal et al. 1988, 1993, 1996; Katzman et al. 1988; Dickson et al. 1992; Mufson et al. 1999). Therefore, abundant Aβ SPs are necessary but not sufficient for a diagnosis of AD. However, Cummings and Cotman (1995) suggest that β-amyloid is central to the etiology of AD. Revised criteria recommended by the National Institute on Aging (NIA)–Reagan Institute base the diagnosis of definite AD on the presence of abundant neurofibrillary tangles (NFTs) as well as SPs in the brain of a demented patient (Anonymous 1997). In contrast to SPs, a number of studies have reported that the progression of cognitive impairment in AD correlates with the abundance of NFTs in the frontal cortex (Di Patre et al. 1999), hippocampus (Caramelli et al. 1998; Haroutunian et al. 1999), and entorhinal cortex (Haroutunian et al. 1999). Nevertheless, the degree to which each of these hallmark features contributes to the dementia of AD remains unclear.

In addition to NFTs, tau-rich neurofibrillary alterations of AD include neuropil threads (NTs) and SP-associated dystrophic neurites (Goedert et al. 1997). These lesions are most prominent in select subfields and layers of the ventromedial temporal lobe and association cortices (Arnold et al. 1991; Braak and Braak 1991; Arriagada et al. 1992a, b; Cotman and Anderson 2000). Like NFTs, NTs contain paired helical filaments (PHFs) composed of hyperphosphorylated tau (PHF-tau), but they occur in axons and dendrites instead of in neuronal perikarya (Schmidt et al. 1993; Goedert et al. 1997).

In addition to the AD brain (Braak and Braak 1988; Ihara 1988; Braak et al. 1994; Komori et al. 1997), NTs also are prominent in the brains of patients with other tauopathies, such as corticobasal degeneration (Matsumoto et al. 1996; Takahashi et al. 1996; Komori et al. 1997; Schneider et al. 1997), progressive supranuclear palsy (Hauw et al. 1990; Braak et al. 1992; Li et al. 1996; Komori et al. 1997), and autosomal dominantly inherited frontotemporal dementia with parkinsonism linked to chromosome 17 (FTDP-17) (Spillantini et al. 1998). Although NTs have been analyzed (McKee et al. 1991; Markesbery et al. 1993) and often referenced in neuropathology reports, a quantitative methodology using a random systematic sampling and image analysis of these neuritic pathologies, i.e., the NT burden, has not been published.

Quantitation of the NT burden is critical in clinicopathological correlative studies of the relative contribution of NTs to measures of the clinical manifestations of neurodegenerative dementias. For this reason, we developed a protocol for the quantitation of NTs in the postmortem brains of elders participating in a large-scale, prospective longitudinal study of cognition in normal and pathological aging. Here we describe a fractionation methodology for sampling brain tissue as well as an image analysis technique that sensitively and reliably quantifies NTs. On the basis of data presented here, we conclude that our protocol enables accurate assessment of the NTs burden in the normal aging brain as well as in the brains of elders with mild cognitive impairment or AD. Moreover, this approach should enable the quantification of normal and pathological neurites labeled by various means in the developing human brain, other brain disorders, and in animal models of these disorders.

Materials and Methods

Subjects

This study included elders with no cognitive impairment (NCI; n = 14), and cognitive impairment (CI), i.e., elders with mild cognitive impairment (n = 9) or dementia due to AD (n = 8; see Table 1). All individuals were participants in the Religious Orders Study, a large, prospective, longitudinal clinicopathological investigation of aging and dementia in Catholic nuns, priests, and brothers. Details of the clinical evaluation procedures have been previously reported (Bennett et al. 1997, 1999; Gilmor et al. 1999; Mufson et al. 1999). Briefly, about 750 elders (>65 years old) from 30 communal groups located in 15 states are currently enrolled in this study. Each participant agreed to an annual detailed clinical evaluation and brain donation at the time of death. The study was approved by the Human Investigation Committee of Rush–Presbyterian-St Luke's Medical Center. Follow-up participation for the annual evaluations exceeds 95% of survivors, and the autopsy rate is over 90%. The average interval from last evaluation to brain autopsy currently is less than 7 months.

Table 1

Demographics

	NCIa	CIb
Number	14	17
Age (mean ± SD)	82 ± 8 yr	85 ± 5 yr
M:Fc	9:5	9:8
Education (mean ± SD)	19 ± 4 yr	16 ± 4 yr
MMSE (mean ± SD)d	27.1 ± 2	24.0 ± 4
PMI (mean ± SD)e	6.1 ± 3 hr	6.5 ± 4 hr

^aNo cognitive impairment.

^bCognitive impairment.

^cMale:female ratio.

^dMini-Mental State Exam.

^ePostmortem interval.

Clinical Evaluation

The clinical evaluation was designed to establish the presence or absence of impaired cognition, including dementia, and to determine the etiology of these impairments, with particular attention paid to common conditions, including AD, stroke, Parkinson's disease (PD), and depression. Briefly, a team of investigators led by a neurologist performed annual uniform, structured evaluations of each participant. The medical history included questions about cognitive decline, stroke, PD, head injury, tumor, depression, and other medical problems, as well as medications used by the participant within the previous 14 days of the examination. Neurological examination was performed by trained nurse clinicians and included an assessment of signs of stroke and parkinsonism. Trained neuropsychology technicians administered a battery of cognitive tests chosen to measure a range of cognitive abilities, with emphasis on those affected by aging and AD. These tests included the Mini-Mental State Examination (MMSE) and other tests recommended by CERAD (Morris et al. 1989; Welsh et al. 1992). Many of these tests were modified slightly to increase portability and to facilitate scoring and record-keeping using a computer (Beckett et al. 1997). After death, all clinical data were reviewed by a team of neurologists and neuropsychologists to arrive at summary diagnoses. The diagnosis of dementia and AD followed the recommendations of the joint working group of the National Institute of Neurological and Communicative Disorders and the Stroke and the Alzheimer's Disease and Related Disorders Association (NINCDS/ADRDA) (McKhann et al. 1984). Although at present there are no consensus criteria for the clinical classification of mild cognitive impairment (MCI), as previously reported, (Gilmor et al. 1999; Mufson et al. 1999), a study has been published that was designed to provide a quantitative characterization of the clinical criteria for the diagnosis of MCI (Petersen et al. 1999). Reproducible and reliable criteria for the clinical diagnosis of MCI would be valuable because of interest in early intervention trials.

Tissue Preparation and Pathological Evaluation

At the time of autopsy, brains were removed and processed as described previously (Mufson et al. 1999). Neuropathological examinations were conducted by board-certified neuropathologists (EJC, JAS) to establish a diagnosis of normal or diseased brain and to identify any other lesions (e.g., stroke) that might confound our analysis (Bennett et al. 1993; Gilmor et al. 1999; Mufson et al. 1999). Each brain was cut into precise 1-cm-thick coronal slabs, immersion-fixed in 4% paraformaldehyde for 24–48 hr, and then placed in 2% dimethylsulfoxide (DMSO)/10% glycerol in PBS for several days. Individual slabs that contained perirhinal and entorhinal cortices and hippocampus were cut and parcelled into a sequential series of 18 sections, each 40 μm thick, on a freezing sliding microtome and then placed back in the cryoprotective solution (Mufson et al. 1999). Sections that contained the intermediate entorhinal cortex (subfield EI), as determined by cytoarchitectural analysis of Nisslstained sections of an adjacent series, were selected for immunohistochemistry and analysis (Insausti et al. 1995).

Immunohistochemistry

Free-floating cryoprotected sections were washed in 0.1 M phosphate buffer (PB). Endogenous peroxidase activity was quenched with 3% hydrogen peroxide in 0.1 M PB for 30 min and then washed in a Tris–NaCl–Triton (TNT; 0.2 M NaCl, 0.5% Triton, 0.1 M Tris) solution. Next, sections were preincubated for 1 hr in 10% horse serum in TNT. Thereafter, tissues were processed by the avidin–biotin method using the Vectastain ABC elite kit (Vector Laboratories; Burlingame, CA). Sections were incubated for 72 hr at 4C with PHF-6 (source: V.M.-Y. Lee), a monoclonal antibody directed against a defined epitope containing phosphorylated T231 residue in PHF-tau (Hoffmann et al. 1997), which has been shown to robustly stain various tau lesions (Reed et al. 1998; Mirra et al. 1999; Wang et al. 1999). Incubation of sections with this antibody was performed at a dilution of 1:20,000 in 10% horse serum in TNT. Control sections processed without the PHF-6 antibody displayed no specific staining. After the incubation with PHF-6 was complete, sections were washed in 0.1 M TBS, incubated with biotinylated secondary antibody (biotinylated anti-mouse IgG in 10 ml of 10% horse serum in TNT) for 1 hr at room temperature, and then washed again in 0.1 M TBS. To enhance the reaction product, sections were preincubated for 10 min in an imidazole acetate buffer [5% 0.2 M imidazole (Sigma; St Louis, MO), 5% 1.0 M sodium acetate (Sigma) in distilled H₂O, pH 7.2] and then reacted for 5–7 min in a solution of 50 ml imidazole acetate buffer with 25 mg 3,3′-diaminobenzidine (DAB) and 1.25 g nickel sulfate (Sigma) with 150 μl of 1% H₂O₂. The reaction was stopped by washing the sections in 0.1 M PB. Sections were mounted on gelatin-coated slides, dehydrated in a series of ethanol washes, cleared in xylene, and coverslipped using Permount (Fisher Scientific; Pittsburgh, PA).

Immunodetection and Quantitation

Individual coronal sections were chosen from series that contained perirhinal (transentorhinal) and entorhinal cortex at the intermediate level of subfield EI and these regions were delineated at low magnification according to cytoarchitectural criteria (Insausti et al. 1995). The basis of our sampling strategy was to obtain an estimate for which the precision was maximized by measuring NFTs and NTs by variable amounts of sampling. Our sampling scheme was organized into a hierarchy that included a group of individuals and, below this, a single section, and then multiple disectors, using standard stereological rules (West et al. 1991). Using this sampling method, a systematic random sampling scheme was performed to quantify the density of NFTs and, with the same sampling pattern, to obtain images for measurement of NT burden (% area). The perirhinal and entorhinal NFT and NT quantifications were performed with a stereological mapping station which included a Leica DMRBE microscope equipped with a ×20 objective, a ×1 phototube, and connected to a camera lucida-like microscope/computer interface (Lucivid; MicroBrightField, Colchester, VT), and a computer (Millennia Mme; Micron Electronic, Nampa, ID) equipped with Windows 95 software (Microsoft; Redmond, WA) and StereoInvestigator software version 3.23 (MicroBrightField).

To analyze the individual images for NT burden, 24-bit color images obtained from the random systematic sampling (40–60 images per perirhinal cortex or entorhinal cortex) were converted to 8-bit gray scale images by using conversion algorithms associated with Adobe Photoshop 5.02 (Adobe Systems; Tucson, AZ). Area analysis was performed using the public domain Object-Image 1.62p15 (developed by Norbert Vischer, http://simon.bio.uva.nl/objectimage.html). Overall data reduction yielded a value for the area occupied by the immunopositive zone or the area fraction (burden) defined as the percent fraction of labeled to unstained tissue area.

The analysis algorithm segmented each image along the intensity domain into two fractions, the labeled and the background compartment. To do so reliably, the system arbitrated between two automatic thresholding algorithms, ISODATA (Iterative Self-Organizing Data Analysis) and triangulation (Bezdek 1980). Both approaches are histogram-dependent procedures. The former seeks to maximize the distance between the mean gray value of two hypothesized compartments (Bezdek 1980), while the latter searches for a minima in the histogram. The former is effective when labeling is extensive and the latter when it is low. It was empirically determined that, during threshold setting using the triangulation procedure, it was possible to flag failure due to too large an immunolabeled compartment. The logic in the triangulation algorithm is illustrated in Figure 1. When the immunopositive compartment is too large, the histogram is shifted and no local minima is detected.

The thresholding step separates from the background PHF-tau immunoreactive structures that are a combination of both NFTs and NTs. A size filter then separates out NFTs from this segmented compartment. The resulting binary image can then be edited manually to reduce false-positives and -negatives.

Figure 1

Automatic thresholding using triangulation. The procedure relies on the image histogram (A), the distribution of pixel gray values within the image. Both the peak and the darkest gray value within the histogram are identified and a line connecting the two is defined (note that 0 gray value corresponds to white and 255 to black). The distance between that line and the underlying histogram is measured. (B) A graph illustrating that distance as a function of gray value is shown over the gray value, ranging between maximum and darkest value of the histogram. The gray value at the peak of this function defines the threshold setting (arrow in B). The corresponding position on the image histogram is demarcated by a dashed line in A.

The sensitivity of this measure to the camera settings employed during image acquisition system was handled by strictly adhering to a fixed protocol. All microvideographs were captured using a Cohu 1300 series color CCD camera (Cohu/Electronic Division; San Diego, CA) with the gamma set to 1.0 and the autogain off. The analysis results are a function of illumination intensity as described in the results (lamp setting on the microscope is not a reliable indicator of illumination level). To assure consistent light level from one imaging session to the next, a calibration protocol was employed. Before each imaging session, the Leica DMRBE microscope was calibrated under KÖhler illumination using an optical density standard (Kodak Step Tablet #152 3422), 0.19 OD, which was cut and attached to a microscope slide. The standard was imaged using ImageJ (National Institutes of Mental Health, available on the Internet at http://rsb.info.nih.gov/ij/). ImageJ calculates and displays a histogram of the distribution of gray values along the x-axis. The light illumination is adjusted until the mean value of the image histogram equals a predefined value, which in these studies was 85 (scale 0–255).

Statistics

Statistical analysis included ANOVA, post-hoc testing, and rank correlations (Spearman's rank correlation). All testing was performed with StatView version 5.01 (SAS Institute, www.statview.com).

Results

Image Analysis

We assessed the performance of our image processing approach and its sensitivity to illumination intensity. There was a dramatic effect of light level on analysis results (Figure 2). As light level increased, the measured NT burden decreased. Although the absolute burden is not relevant, it is critical that a consistent protocol is followed when all data of a given study are imaged. To standardize analysis, we selected a level that yielded a gray value of 85 (range 0–255) when the 0.19 optical density (OD) calibration standard was imaged.

Figure 2

Variation in percent neuropil PHF-tau immunoreactivity (antibody PHF-6) quantification based on illumination values. As the microscope illumination is increased, there is a reduction in percent area fraction occupied by the PHF-tau-immunopositive filament. Abscissa values refer to the control setting on the microscope for illumination intensity. Using NIH ImageJ software, the mean gray value of the calibration standard ranged from 60 gray value (bin defined as 5.5) to 240 gray value (bin defined as 8.0) on a scale ranging from 0 to 255. Using a t-test (significance level set at p >0.05), x-axis value of 5.5 illumination setting was significantly different than all other x-axis values: 6.0 was significantly different from 7.5 and 8.0; 6.5 was significantly different from 7.5 and 8.0; 7.0 was significantly different from 7.5 and 8.0; 7.5 was significantly different from 8.0.

Figure 3

Repeatability of image capture methodology. We chose three brains that represented one case each of mild, moderate, or marked entorhinal PHF-tau immunoreactivity. From each of the three brains chosen, three fields of view from the entorhinal cortex were imaged. The exact same fields were imaged 4 and 6 months later (April, black bars; August, light gray bars; October, dark gray bars). Each set of bars shows measurements from one brain at three time points. The PHF-tau immunoreactivity was calculated and then compared to evaluate the variability in percent area obtained among the repeated calibration and measurement runs.

By employing a standardized image acquisition protocol, images obtained on different days can be compared. We repeatedly captured three random images from the entorhinal cortex of cases with either low, moderate, or marked NT burden over a period of 6 months. To ensure that the specific location of each image for temporal comparison was identifiable, the exact coordinates of each of the images was determined using the 3-axes motorized stage. Images were then analyzed for NT burden as described above. No substantial difference was detected (Figure 3).

We examined false-positive/false-negative rates on images using the standard protocol. The images were obtained from cases with a low level of NT burden (n = 1), a moderate NT burden (n = 2), and marked NT burden (n = 1; Table 2). Examples of images from those cases and the segmentation results are shown in Figure 4. To examine the performance rate, the binary images generated by the program were further manually edited. The NT quantification accuracy varied depending on the thread burden, with 3.4% false-negatives when the thread burden was mild, 9.8% when the thread burden was moderate, and 9.7% when the thread burden was marked (Table 2). More importantly, there was a 0.0% false-positive measurement regardless of the thread burden.

We also determined how the percent contribution of NTs compares to the overall PHF-tau immunoreactivity which includes NFTs. In images of perirhinal and entorhinal cortex from three participants (low, moderate, and marked PHF-tau immunoreactivity), NT burden contributed 91.4 ± 2.5% (n = 6; range 87.3–93.9%) of the PHF-tau immunoreactivity, and NFTs contributed only 8.6 ± 2.5% (n = 6; range 6.1–12.7%). Even in a case with a high NFT density (44.6 NFTs/mm² in the perirhinal cortex and 21.7 NFTs/mm² in the entorhinal cortex), the NFT contribution of PHF-tau immunoreactivity was >13%.

Sampling Paradigm for NTs and NFTs

For NTs and NFTs, we identified a participant whose brain PHF-tau burden was considered to be moderate with a heterogeneous distribution. Throughout the contour sampled (32 × 10⁶ μm²), we positioned an unbiased counting frame of known area, 40,000 μm², along a rectangular grid (Table 3) superimposed on the perirhinal and entorhinal cortex by the StereoInvestigator software. The coordinated movement along the x-axis and y-axis was achieved by StereoInvestigator software integration, with stepping motors attached to the microscope stage. For the NT burden measurements, we sampled the entorhinal cortex of a moderate NT burden case with a heterogeneous distribution and a mild NT burden case. Three sets of images were obtained from each case. The moderate thread burden case was examined by using a systematic random sample of 20 images, 40 images, and 60 images (Table 3). A total of 40–60 images or 25–35% of the area must be sampled to obtain actual values of thread burden. A mild thread burden case (Table 3) was examined by using a systematic random sample of 20 images, 30 images, 40 images, and 50 images. It is apparent that, regardless of thread burden, attention must be directed towards adequate sampling because the NT burden is negatively correlated with sampling when there is a heterogeneous distribution.

Table 2

Error detection in neuropil thread analysis a

A.	Thread burden	True-positives	False-positives	False-negatives	% False-positives	% False-negatives
	Mild	7194	0	702	0.0	9.8
	Moderate	62,645	0	6,105	0.0	9.7
	Marked	118,856	0	4,024	0.0	3.4
B.	Thread burden	True-positives	False-positives	False-negatives	% False-positives	% False-negatives
	Moderate	31,080	20,465	17	65.8	0.1

^a Table 2A data were obtained using a microscope illumination setting of 5.5 (mean gray scale of 85). Table 2B data were obtained using a microscope illumination setting of 6.5 (mean gray scale of 175). Values listed for neuropil thread burden are area in pixels.

Figure 4

Videomicrographs of neuropil threads (NTs) (A,C,E) and corresponding binary images after image analysis as described in Materials and Methods (B,D,F). Images are from three cases with a light (A,B), moderate (C,D), and marked (E,F) NT burden in the entorhinal cortex. All panels are at the same magnification. Bar = 50 μm.

Table 3

Sampling analysis for NFTs and NTs

A.	Sampling boxesa	Random sets	Area measured (%)	NFTs/mm2 (± SE)	Sampling box grid (X, Y)–μmb
	100	n = 9	11.4	43.5 (2.1)	870; 490
	50	n = 11	6.6	41.7 (2.6)	1200; 570
	25	n = 10	3.1	29.9 (3.0)	1730; 830
B.	Burden	Imagesa	Area measured (%)	Neuropil thread (NT) burden (%)
	Moderate	60	32.2	34
	Moderate	40	23.0	35
	Moderate	20	10.4	30
	Mild	50	13.7	18
	Mild	40	13.0	15
	Mild	30	13.8	11
	Mild	20	13.5	7

^aEach box or image was measured as 405 μm × 308 μm, for a total image area of 124,740 μm². The relationship between images obtained for NT burden analysis and NFT density is described in Materials and Methods. All NT burden data were obtained using a microscope illumination setting of 5.5 (mean gray scale of 85).

^bThe regions of interest were sampled in a random systematic pattern. The number of sampling boxes (100, 50, or 25) is directly related to the size of the step movements made on the x and y axes, shown in this column.

For NFTs, we determined that a sampling scheme which included at least 50 images (6–7% of total area) was ample to estimate accurately the NFT density and to simultaneously be efficient. By doubling the number of images taken in a systematic random sample (n = 100; Table 3), the change in NFTs was not significantly different (Student's t-test; t = 0.615, p <0.05). In contrast, by reducing the number of images taken (n = 25; Table 3), our NFT density decreased by almost 30%, which was significantly different (Student's t-test; t = 3.145, p >0.01).

Pathology

Perirhinal cortex (Figure 5A) and entorhinal cortex (Figure 5B) NT burdens (percent area) were highly correlated (Figure 6; r = 0.9 (r ² = 0.83), p >0.0001) and increased significantly in persons with CI compared to NCI participants, as expected (t = −2.82, p >0.0065).

Discussion

We developed systematic automated measurement methods that enabled us to determine the NT burden expressed as area fraction occupied in the perirhinal and entorhinal cortices. Neuropil threads, also called curly fibers or dystrophic neurites, are abnormal neuronal processes that contain PHFs formed by PHF-tau (Braak et al. 1986). Both the perirhinal and entorhinal cortical regions are the earliest and most strongly affected areas at which neurofibrillary deposits occur (Arnold et al. 1991; Braak and Braak 1991). Because of the heterogeneous distribution of both NFT and NT burden within an area, sampling becomes a potential and significant component for an accurate estimation. Inadequate sample sites would skew results in regions of the brain at which there are heterogeneous distributions of either NFTs or NTs. Systematic uniform random sampling within each section provided us with a guide to modify the amount of sampling necessary to reduce potential variability of the estimation. This was most evident in controls and participants with greater cognitive impairment and higher burdens of AD pathology.

The increasing clinical and biological interest in dendrite and axonal damage in neurodegenerative disease, e.g., NTs in AD (Newell et al. 1999), corticobasal degeneration (Komori et al. 1997; Komori 1999), and synuclein-positive Lewy bodies and Lewy neurites in both dementia with Lewy bodies (DLB) and PD (Spillantini et al. 1997; Baba et al. 1998; Braak et al. 1999; Hurtig et al. 2000), warrants the development of quick, accurate, and reliable methods for quantifying NFT and NT burden in the aging and AD brain. Beyond PHF-tau NTs, immunohistochemical studies using antibodies to α-synuclein have revealed extensive networks of dystrophic processes, termed Lewy neurites. These Lewy neurites are seen in the brains of patients with sporadic DLB, PD, and the LB variant of AD (Spillantini et al. 1997; Wakabayashi et al. 1997; Baba et al. 1998), and our methods would be suitable for their analysis too.

Accurate estimation of neuropil filament burden requires stereological analyses as much as does evaluation of neuron or synapse number (West et al. 1991; West 1999). If the marker of interest is uniformly distributed in a homogeneous fashion, sampling can be minimal and the expected density or absolute count should reflect the actual value. However, as is often the case for neuron distribution, the pattern of pathological lesions within distinct regions of the brain is typically nonhomogeneous. In this study, perirhinal and entorhinal cortices from each participant were used for quantification of PHF-tau-positive structures (NFTs and NTs). Because we selected a single representative section for each case, it is possible that our design-based stereological study is biased in the sense that all NFTs and NTs throughout the entire volume (three dimensions) of the structure of interest did not have an equal probability of being counted. This notwithstanding, we implemented a sampling scheme that was based on systematic, uniform random sampling using the fractionator (MicroBrightField). Because of the large area of the human perirhinal and entorhinal cortex in each section, we wanted to adequately sample to obtain low variation but also to be efficient. We found that the distribution of PHF-tau structures was heterogeneous and thus not predictable and, regardless of the apparent thread burden, sample size must be considered and a critical area needs to be analyzed. At the risk of oversampling, it is suggested that the actual area sampled be in excess of that expected for absolute efficiency. This can be accomplished by increasing the number of sampling sites or area analyzed per individual section and by adding additional systematically obtained sections from the same region of interest. Stereological analyses of neuron numbers in normal aging and AD has produced evidence for some degree of neuronal loss (West 1993, 1994; Gómez-Isla et al. 1997; Simic et al. 1997; Hof et al. 1999). However, pathologies other than absolute neuron loss, e.g., subtle changes to the dendritic arborization, may be more important in the course of disease. The ultimate evaluation of whether absolute values or densities of PHF-tau-positive neurofibrillary structures are meaningful will be determined only with correlations with diagnostic and clinical variables (unpublished data).

Figure 5

Total PHF-tau immunoreactivity (percent area) in the perirhinal cortex (A) and entorhinal cortex (B) of controls (NCI or no cognitive impairment; n = 14), and participants with cognitive impairment (CI; n = 17).

Figure 6

Correlation between perirhinal (PC) and entorhinal (EC) cortex total PHF-tau immunoreactivity of all participants (n = 31). There was a significant rank correlation between the EC and the PC PHF-tau immunoreactivity (r = 0.857, p >0.0001).

Both Markesbery et al. (1993) and McKee et al. (1991) published data from morphometric image analysis of tau lesions in AD. Comparisons with our study are difficult because their image analysis procedures were not described in sufficient detail to enable us to discern similarities and differences. Moreover, Markesbery et al. (1993) quantitated NT levels from the superior frontal gyrus, in contrast to the entorhinal cortex reported here. Studies are under way to measure NFTs and NTs in neocortical regions from a similar cohort of subjects described here, which will enable more meaningful comparisons. McKee et al. (1991) reported on quantitation of dystrophic neurites and NFTs in various neocortical regions, including a similar NFT density to our data from the entorhinal cortex (Table 3A).

The quantification of PHF-tau-positive structures yielded a composite score, defined as a combination of the black pixels that identified both NFTs and NTs. To separate the area contribution of NFTs and NTs, an NIH Image macro was developed to efficiently and accurately delineate and exclude NFTs and produce values that are exclusively defined as NTs. This macro selects (on the basis of contrast) the objects (NFTs) in a ×200 magnification (×20 objective) PHF-tau field, filters out objects too small (individual NTs) or too large (masses of NTs) to be NFTs, and then measures the area of each object in the image. The variable “size cut” determines which objects get counted. Observer editing is then performed on each image to shield against false-positives and -negatives. Although additional data analysis is in progress, our results indicate that NT burden contributes heavily, with 91.4 ± 2.5% of the PHF-tau immunoreactivity, and that NFTs contributed only 8.6 ± 2.5%. The necessity for increased accuracy of NTs is being presently evaluated in clinicopathologic studies.

New guidelines for postmortem diagnosis of AD were established in 1997 (Anonymous 1997) by the NIA–Reagan Institute. The NIA–Reagan Institute criteria use the topographic staging of AD proposed by Braak and Braak (1991) on paraffin sections stained by a modified Bielschowsky silver method to grade the severity of the burden of NFTs and NTs in five brain regions. In a recent study by Newell et al. (1999), the utility and validity of the NIA–Reagan Institute criteria in comparison to those of the Khachaturian and CERAD criteria showed a good correlation between the NIA–Reagan criteria and clinical dementia. However, only semiquantitative assessments were used. Quantitative assessments of NT burden will provide a more sensitive measure of dendritic structural damage, and this will be especially useful for correlations with clinical and neuropsychological measures obtained ante mortem.