Brainstem tau pathology in Alzheimer’s disease is characterized by increase of three repeat tau and independent of amyloid β

Twenty three autopsied brains with different cortical NFT stages (Table 1, Braak and Braak’s NFT stages I/II: 8 cases, III/IV: 8 cases, V/VI: 7 cases) were enrolled in this study with informed consent and the approval of the ethics committee of Tokyo Metropolitan Institute of Medical Science and Nitobe-Memorial Nakano General Hospital. These cases have previously been anonymized and subjected to standardized neuropathological assessment at Tokyo Metropolitan Institute of Medical Science, including silver impregnations by Gallyas and Campbell-Switzer method, and DAB immunohistochemistry with anti-phospho-tau (Ser202, Thr205) antibody AT8 (Thermo Fisher Scientific K.K., Tokyo, Japan) of representative brain regions. The brainstem sections of all cases, and cortical sections of selected cases underwent DAB immunohistochemistry with anti-alpha-synuclein phospho (Ser129) monoclonal antibody (1:10,000, Clone pSyn#64, Wako, Japan). Diagnosis of NFT stage, Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) neuritic plaque score, and the evaluation of Lewy body distribution and frequency were performed following the published criteria [5, 37, 38]. Cases with non-Alzheimer-type tau deposits (e.g. progressive supranuclear palsy, corticobasal degeneration, argyrophilic grain disease and Pick body disease) were excluded. Concomitant Lewy pathology was observed in case 1, 9, 13 and 23 (Table 1, Additional file 1).

The autopsied brains were immersion-fixed in 4% formalin for 4 weeks. The brainstem was sliced perpendicular to the axis and embedded in paraffin. The levels of the sections were coordinated between cases by encompassing landmark anatomical structures (superior colliculus (SC) and red nucleus (RN) for the midbrain, and LC and superior cerebellar peduncles for the pons) (Fig. 1a). The 6-μm thick sections were deparaffinized for double immunofluorolabeling with antibodies against isoform-specific anti-4R tau antibody (rabbit polyclonal, Cosmo Bio Co, Tokyo, Japan), raised against amino acids 275–291 of human 4R tau, which is deamidated at N279 [10, 22], and anti-3R tau antibody (RD3, mouse monoclonal, Merck Millipore, Germany) [11]. To reduce diffuse cytoplasmic staining with the RD3, the sections were pretreated with 0.25% potassium permanganate for 15 min, 2% oxalic acid for a minute, and >99% formic acid for 30 min at room temperature, and autoclaved in 0.05 M citrate buffer for 20 min at 120 °C [53]. This pretreatment also improved immunolabeling with polyclonal anti-4R tau (data not shown). Sections were washed with phosphate-buffered saline (PBS) with 0.03% polyoxyethylene (10) octylphenyl ether (Triton X-100), blocked for 30 min in 5% normal goat serum/0.05% NaN₃/PBS with 0.03% Triton X-100, and incubated with polyclonal anti-4R-tau antibody (1:3000) and RD3 (1:300), diluted in the blocking buffer at 4 °C for 4 days. To reduce autofluorescence of lipofuscin, sections were treated with Sudan Black B [44]. Quenching of the autofluorescence of lipofuscin was confirmed by spectral analysis (Additional file 2: Figure S1). Primary antibodies were labeled with Alexa 488 conjugated with anti-rabbit IgG (Molecular Probes, Oregon, USA, 1:200) and Alexa 568 conjugated with anti-mouse IgG (Molecular Probes, Oregon, USA, 1:200), respectively, diluted in PBS with 0.03% Triton X-100 overnight in the dark. Sections were mounted with buffered glycerol containing 0.1% p-phenylenediamine. For anatomical assessment, immediately adjacent sections were stained with Hematoxylin & Eosin and Klüver-Barrera (KB) methods.

For precise comparison of Aβ with tau deposits, neighboring sections from the same tissue blocks used for tau immunohistochemistry were subjected to Aβ/DAB immunohistochemistry. For the confirmation of the observation of 4R and 3R tau double-immunofluorolabeling, representative sections were also subjected to RD4 (mouse monoclonal anti-4R tau antibody, Merck Millipore, Germany) or RD3/DAB immunohistochemistry [11]. For Aβ, the sections were treated with > 99% formic acid for 5 min at room temperature. For RD4 or RD3, sections were pretreated as described in the section above [53]. After being treated with 1% H₂O₂ for 30 min, the sections were incubated in primary anti-Aβ mouse monoclonal antibody (Clone 6F/3D, M 0872; DAKO, Agilent Technologies, California, USA, 1:1000) at 4 °C for 2 days, RD4 (1:1000), or RD3 (1:3000) at 4 °C for 4 days, and biotinylated secondary anti-mouse antibody (BA-2000, Vector, California, USA, 1:1000) for 2 h. After subsequent incubation with streptavidin biotinylated horseradish peroxidase complex (ABC Elite, Vector, California, USA), color development was performed with DAB in the presence of imidazole and nickel ammonium chloride. They were examined by light microscopy, and the same virtual slide system used for tau deposits. For semi-quantitative grading of Aβ deposition, only the number of well-defined, compact Aβ deposition was counted, while ill-defined, fleecy deposition and dot-like deposition (smaller than approximately 1μm²) was not included in the count. The amount of well-defined, compact Aβ deposition in 1 mm² microscopic field with maximum deposition density was graded as follows; − (None): no amyloid deposition in the field, + (Sparse): 1–9/field, ++ (Mild): 10–19/field, +++ (Moderate): 20- /field (Table 1).

We obtained a virtual slide image covering full range and depth of the section at high resolution, and enrolled all the immunofluorolabels in the whole investigated area in comprehensive particle analyses. For this, we used a virtual slide system VS120 (Olympus, Tokyo, Japan) equipped with UPLSAPO 10 times objective lens (Olympus, Tokyo, Japan), a triple-band dichroic mirror unit (U-DM3-DA/FI/TX, Olympus), band path filter with peak emission wavelength at 518 nm (for 4R tau/Alexa 488) and 580 nm (for RD3/Alexa 568), a sensitive cooled charge-coupled device camera (ORCA-R² C10600-10B, Hamamatsu Photonics, Shizuoka, Japan), and VS-ASW software (Olympus, Tokyo, Japan). Serial snapshots of a double-immunofluorolabeled section (4R tau/Alexa 488 and RD3/Alexa 568) were captured by VS120 on motorized stage with 10 times objective lens in separate fluorescence channels, and put together to make a seamless broad image, covering the whole tectum and tegmentum (Fig. 1a, Additional file 3: Figure S2). The resolution of each snapshot was 1376 pixels (horizontal) × 1038 pixels (vertical) at 0.645 μm/pixel with 10 times objective lens (original 8 bit). After subtracting the overlapping margins, the area per snapshot reduced to approximately 1138 pixels × 834 pixels (0.398 mm²/snapshot). 5 vertical planes at 1 μm intervals were simultaneously captured (Fig. 1a). To show all the immunolabels at full depth of the section with high accuracy, the EFI program on CellSens software (Olympus, Tokyo, Japan) extracted the pixels with maximum local contrast from the 5 vertical planes and made a single in-focus image (Fig. 1b, c-e), which minutely depicted even small threadlike lesions. These images were converted to big-tagged image file format, retaining the original resolution, to be further operated on ImageJ (NIH, Bethesda, Maryland, USA). The images were uniformly binarized in separate channels according to the threshold operationally defined by Triangle algorithm [59] on ImageJ (Fig. 1f, g). Colocalization analysis between binary images was subsequently performed with ImageJ plug-in (P. Bourdoncle, Institut Jacques Monod, Paris, France) (Fig. 1h). Particle analysis program of ImageJ showed particle size of each immunofluorolabel automatically, as well as X-Y coordinate, in separate fluorescence profiles. We defined particles with a 1–200 μm² area as NTs (Fig. 1i-k), and with an area larger than 200 μm² as NFTs (Fig. 1l-n). The results of the analyses were inspected against the original counterparts to ensure that each lesion was properly represented, and apparent contaminants (e.g. nonspecific staining of the vessel walls) were excluded from the count. For analysis of the regional differences, necessary parts were extracted from the virtual slide images. The regional counts in the 1 mm² field with maximal NFT density was graded as follows; − (absent): no NFTs in the field, + (sparse): 1–9/field, ++ (mild) 10–19/field, +++ (moderate): 20- /field (Table 1). The results of the particle analyses were highlighted as regions of interest (ROIs) on ImageJ. The outline width of ROI depiction was uniformly widened and flattened onto a blank image file with same pixel size as the original. These abstracted outlines served as immunolabel mappings of NTs (Fig. 1o-q) and NFTs (Fig. 1r-t) in different tau-isoform profiles. Also, the virtual slide images of Aβ/DAB immunohistochemistry were captured with 10 times objective lens, and underwent thresholding by RGB values (R: 0–110, G: 60–150, B: 60–150) on CellSens software (Olympus, Tokyo, Japan). This procedure separated black-brown DAB labeling from reddish-brown neuromelanin and particle analysis highlighted the former as ROIs. The outline widths of these ROIs were uniformly widened and simplified on ImageJ to serve as the immunolabel mapping of Aβ deposition. We propose this virtual-slide based quantitative analysis of multifluorolabeled specimen as “Complete ENumeration and Sorting for Unlimited Sectors (CENSUS)” via immunofluorescence.

All statistical analyses were performed using software R (version 3.2.3, R Foundation for Statistical Computing, Vienna, Austria.). P < 0.05 was considered significant. To find the differences in the means of data between cases, Fisher’s exact test or paired t-test with Bonferroni correction was performed. The counts of NTs and NFTs in each tau isoform, and their proportion to the total counts on the entire section were obtained. The proportional data underwent arcsine transformation prior to further analysis to make the data distribution closer to normal. Since arcsine of 1 equals 1.57, axis for the plot of arcsine-transformed proportion ranged from 0 to 1.57. The characteristics of individual cases were stratified into 3 subgroups as follows; age at death: young-old (ages 63–80 years, n = 7), middle-old (81–90 years, n = 9), and very-old (91–102 years, n = 7), NFT stages: I/II (n = 8), III/IV (n = 8) and V/VI (n = 7), CERAD plaque score: 0/A (n = 9), B (n = 8), C (n = 6), and the formalin-fixed brain weight (g): above 1350 (n = 8), 1200–1350 (n = 7), and below 1200 (n = 8). To assess whether there is a significant orderly increasing or decreasing trend along these stratifications, Jonckheere’s trend test was performed. To explore the best-fitting regression models of the plots, the model that yields the least Akaike’s Information Criterion was searched from quadratic (y = ax ²  + bx + c), linear (y = bx + c or y = c), exponential (y = ab ^x ), and power (y = ax ^b ) regression models (the formulas for all the regression models are available on request). The NFT stages (I/II, III/IV, V/VI), CERAD neuritic plaque score (0, A, B, C) and brainstem Aβ deposition (−, +, ++, +++) were converted to numeric variables from 0 to 3 [43]. Odds ratio and 95% confidence intervals for the percentages of sections with greater proportion of 3R tau than 4R tau during the advancement of cortical NFT stages (from I/II, III/IV to V/VI) were calculated by a univariate binomial logistic regression model (Additional file 1). In the box plots, the box represents the 25th and 75th percentiles, the horizontal line in the box represents the median, whisker show the 10th and 90th percentiles, and the white circles represent the outliers. Lines with asterisk indicate statistically significant differences.

Midbrain and pontine sections from 23 cases with different NFT stages were double-immunofluorolabeled for 4R and 3R tau (Table 1). The neurofibrillary changes were present in midbrain and pons of all the investigated cases (Table 1). We captured 300,860 snapshots (approximately 0.398 mm² per snapshot after subtracting the overlapped area) of these sections in total, which were 30,086 snapshots on XY planes with 5 vertical planes in 2 fluorescence channels, by the virtual slide system (detailed in Additional file 1). The investigated area was approximately 120 cm² in total. We performed comprehensive quantitative analyses on these images as described above (Fig. 1). The data sets obtained from approximately 286 gigabytes of image files consisted of 847,763 NTs (602,839 in the midbrain, 244,924 in the pons) and 7859 NFTs (4948 in the midbrain, 2911 in the pons). Each data retained the information of the location and tau isoform profile of the particle. Based on the results of quantitative analyses, immunolabel maps of the NTs and NFTs in different tau isoform profiles were operationally drawn (representative maps in Fig. 2).

The topographical distribution was similar between NFTs and NTs (Fig. 2, bottom two rows of midbrain and pons), and between 4R and 3R tau (Fig. 2, left four columns). The neurofibrillary changes were accentuated at the periaqueductal gray (PAG, ventral half-dominant), linear raphe nucleus (LRN), ventral tegmental area, and substantia nigra (SN) of the midbrain, and the LC, DRN, median raphe nucleus (MRN), and parabrachial nucleus of the pons (Fig. 2). Immunolabel maps of Aβ were also operationally drawn (Fig. 2, right column). Aβ deposits were accumulated at the SC, PAG (dorsal half-dominant), SN (medial half-dominant), LC, reticular formation (RF), and MRN. Topographical differences between tau and Aβ deposits will be described later in detail.

We assessed whether the total counts of neurofibrillary changes in the brainstem increase with disease progression (Fig. 3a a-d, b a-d). Jonckheere’s one-sided test for increasing trend generally revealed significant increasing trends of the total counts of NTs and NFTs, with advancing age and Alzheimer-related cortical pathologies; i.e., cortical NFT stage, CERAD neuritic plaque score, and brain weight reduction (Fig. 3a a-d, b a-d, p < 0.05), apart from a minor exception (Fig. 3a d, the count of the midbrain NTs against brain weight reduction, Jonckheere’s trend test, p = 0.054).

To see which of the tau isoforms contributes to the increase of total counts of neurofibrillary changes with disease progression, we calculated the proportion of 4R and 3R tau-positive neurofibrillary changes to the total, and compared their overall means by paired t-test (Fig. 3a e, k, b e, k). The overall mean of the proportion of 3R tau-positive pontine NFTs (Fig. 3b k, red box plot) was significantly higher than that of 4R tau (Fig. 3b k, green box plot, paired t-test with Bonferroni correction, p = 0.0093), while the difference was not significant for the NTs of the midbrain and pons (Fig. 3a e, k) and the NFTs of the midbrain (Fig. 3b e) (paired t-test with Bonferroni correction, p = 1.0, 1.0 and 0.68, respectively).

We next assessed whether the proportion of 4R and 3R tau-positive NTs and NFTs alter with disease progression at the midbrain and pons. Jonckheere’s one-sided test for increasing or decreasing trend (detailed in method) revealed that the NTs of the midbrain and pons and the NFTs of the midbrain showed significant increasing trends in the proportion of 3R tau-positive lesion with advancing age (Fig. 3a f, l, b f), NFT stage (Fig. 3a g, m, b g), CERAD neuritic plaque score (Fig. 3a h, n, b h), and reducing brain weight (Fig. 3a i, o, b i). Furthermore, in the subgroups of very-old age (age 91-), advanced NFT stages (V/VI) and low brain weight (below 1200 g), paired t-test revealed that the mean of the proportion of 3R tau-positive midbrain NFT was significantly higher than that of 4R tau (paired t-test with Bonferroni correction, p = 0.0017, 0.033, 0.0017, respectively, detailed in Additional file 1), indicating that the proportion of 3R tau-positive midbrain NFTs progressively increases and dominates over that of 4R tau in the advanced phase. The proportion of 4R/3R tau both-positive lesion also showed significant increasing trend for NTs of the midbrain and pons (Fig. 3a f-j, l-p) and the NFTs of the midbrain (Fig. 3b f-j), with advancing age and Alzheimer-related pathologies, with a minor exception (midbrain NFT against brain weight reduction, Fig. 3b i, Jonckheere’s trend test, p = 0.061).

On the other hand, the pontine NFTs did not show any increasing or decreasing trend for the proportion of 3R tau or 4R/3R tau both positive lesions (Fig. 3b l-p) on Jonckheere’s trend test. Together with the result of paired t-test (Fig. 3b k), this result indicated that the proportion of 3R tau-positive pontine NFTs were persistently elevated from younger age and early phase of the disease.

Taken together, these findings indicated that, along the increase in the total count of neurofibrillary changes, the proportion of 3R tau-positive midbrain NFTs increased with disease progression and became significantly higher than that of 4R tau in advanced disease subgroups, while the proportion of 3R tau-positive pontine NFTs remained stably elevated from the early phase.

The adjacent sections taken from the same tissue blocks as the representative sections depicted in Fig. 2 were subjected to single DAB immunohistochemistry with RD4 or RD3 antibodies to assess the reproducibility of the observation by immunofluorescence. DAB immunohistochemistry of these representative sections was in line with the finding based on immunofluorescence that the proportion of 3R tau positive NFTs in the midbrain increased with advancing NFT stages, while the proportion of 3R tau positive NFTs in the pons was high already in early stages.

We next aimed to clarify the topographical distribution of neurofibrillary changes on the midbrain and pontine sections from 23 cases (Table 1, Fig. 5). In 80% of the cases with NFT stages III-VI (Table 1, 12 out of 15 cases), neurofibrillary changes affected the ventral half of PAG more severely (Fig. 5a, b, open arrows) than the dorsal counterpart (Fig. 5a, b, solid arrows, p = 0.000011, Fisher’s exact test). By comparison to KB-stained section (Fig. 5d, nearby section of Fig. 5b), the ventral predilection sites of neurofibrillary changes at the PAG were rich in large neurons of DRN (Fig. 5d, open arrow), while the dorsal counterparts were rich in smaller neurons (Fig. 5d, solid arrow). At the midbrain (Fig. 5a, Table 1), the neurofibrillary changes were also accentuated at the LRN (Fig. 5a, open arrowhead), ventral tegemental area, and SN (Fig. 5a, solid arrowhead). At the pons, the neurofibrillary changes were accentuated at the LC (Fig. 5c, open arrows), DRN (Fig. 5c, open arrowheads), MRN (Fig. 5c, solid arrowheads), and parabrachial nucleus (also Table 1).

Next, we compared the topographical distribution of Aβ deposits to that of tau on neighboring sections at the midbrain and pons. Aβ deposits at the midbrain (Fig. 6a-d) and pons (Fig. 6e-g) showed various morphologies. Fleecy amyloid deposits [50] surrounded some of the capillary walls (Fig. 6d). The amount of Aβ deposits was associated with increasing trends of NFT stages (Jonckheere’s trend test, p = 0.00018 and 0.00037, for midbrain and pons, respectively, Fig. 6h). However, only 57% of investigated cases showed Aβ deposits at either or both the midbrain and pons (Table 1, Fig. 6i, 0% of the cases with NFT stages I/II, 87.5% of stages III/IV and 86% of stages V/VI). For example, case 21, with cortical NFT stages V/VI and CERAD neuritic plaque score 0 (Table 1), exhibited AT8-positive neurofibrillary changes at the midbrain and pons (Fig. 6j, LC), but no Aβ deposits on the neighboring sections (Fig. 6k).

When present, the SC (Fig. 6l, m, open arrowheads) was the most severely affected region of all investigated areas by Aβ deposition (Table 1), and the predilection sites included the LC (Fig. 6n, open arrow), MRN (Fig. 6n, solid arrowhead), and RF (Fig. 6n). Lesions in RN and pontine base were only sparsely detected even in advanced cases. The subnucleus medialis of the PAG (Fig. 6l, m, solid arrowheads), which surrounds the aqueduct, did not show Aβ deposition in the investigated cases.

The predilection sites of Aβ deposition differed considerably from those of tau (Table 1, Fig. 2) at the PAG, DRN, LRN, and SN, as described below. The Aβ deposition affected the dorsal PAG more severely (Fig. 6l, solid arrow) than the ventral counterpart (Fig. 6l, open arrow) in 9 out of 9 cases with any deposition therein (p = 0.000041, Fisher’s exact test). This dorsal predilection of Aβ at the PAG dissociated from the ventral predilection of tau deposits (Fig. 5a, b, open arrows). Also, at the DRN and LRN, where tau deposits were abundant, the Aβ deposits were rarely detected (Table 1). Finally, medial half of the SN was affected by the Aβ deposition more severely than the lateral half in 6 out of 8 cases with any deposits therein (75%, p = 0.0070, Fisher’s exact test), while the amount of tau deposits was not significantly different between medial and lateral SN (p = 1.0, Fisher’s exact test).

Taken together, the Aβ deposition at the midbrain or pons was absent in all the cases with NFT stages I/II, although gradually increased to be present in 86% of stages V/VI. When present, the predilection sites of Aβ deposition included the SC, PAG (dorsal half-dominant), SN (medial half-dominant), LC, RF, and MRN. On the other hand, the ventral half of the PAG, DRN and LRN, which were the areas in which neurofibrillary changes were abundant, were much less intensely affected by Aβ deposition. These differences in distribution suggested that the tau and Aβ deposition may occur independently of one another at least to some extent.

Because our CENSUS approach standardizes the acquisition condition of fluorescence microscopic images and operationally picks up all the immunopositive lesions contained in the histological sections (Fig. 1), the data sets, once obtained, can be arbitrarily processed for size quantification (Fig. 1), mapping based on XY coordinates (Figs. 1, 2), calculations of the proportion of isoform (Fig. 3), and their possible relation to disease progression can be demonstrated (Fig. 3). This CENSUS fundamentally changed conventional morphometric approaches, usually calculating local density of predefined ROIs, to test whether the results corroborate the predefined working hypothesis. Moreover, local density of a lesion may be influenced by the atrophy of the target structure and overall neuronal loss [52]. For example, an increase in the lesion density may be accentuated if the entire structure is atrophied, even when the count number remained relatively unchanged. Because CENSUS approach picks up all the lesions in the section, interpretation is much more straightforward than conventional ROI-based approach.

Previous reports showed that the NFTs matures morphologically from 4R tau dominant pretangles to 3R tau dominant ghost tangles in the hippocampus, and that the proportion of 3R tau-positive neurofibrillary changes was higher in the hippocampal subregions with advanced neurofibrillary pathology than those involved in later stage [21, 28, 34, 35, 54]. Thus, it is hypothesized that this tau-isoform transition during the morphological maturation of the NFT is orchestrated to form the regional progression of 3R tau dominance in the hippocampus along the perforant path containing unidirectional hippocampal circuitry, beginning in the entorhinal and transentorhinal cortices, subsequently progressing to the subiculum and CA1, and further to CA 3–4 [21]. In the previous studies, however, the effect of disease progression on tau isoform prevalence was not fully evaluated [21, 28, 34, 35].

In this study, we enrolled sufficient number of samples to evaluate the effect of disease progression, and by employing CENSUS approach, we clarified that the proportion of 3R tau-positive NFTs in the midbrain and the NTs in the midbrain and pons gradually increased with advancing Alzheimer-related cortical pathology (Fig. 3a f-i, l-o, b f-i). This gradual increase in the proportion of 3R tau probably explain why there was not a significant difference between the overall means of the proportion of 4R and 3R tau in these neurofibrillary changes (Fig. 3a e, k, b e, paired t-test). On the other hand, the overall mean of the proportion of 3R tau-positive pontine NFT was significantly higher than that of 4R tau (Fig. 3b k) and it was stably elevated along disease progression (Fig. 3b l-o, Jonckheere’s trend test), indicating that the proportion of 3R tau-positive NFTs was persistently dominant in the pons but not in the midbrain. This difference does not necessarily indicate that the mechanism of tau deposition is different between pons and midbrain. Rather, this difference is explained if pontine neurons are liable to develop neurofibrillary changes from earlier stage than midbrain neurons. It is then expected that similar progressive dominance in 3R tau may be detectable if pontine samples from younger individuals are included, which is a subject for future studies. If regional gradient of isoform around hippocampus is oriented along a defined major circuity such as performant path, what kind of circuitries in the brainstem are responsible for the gradient? Because neuroanatomical connections in the brainstem are much more complex [4, 39, 40, 42] than that of the hippocampus, it is practically impossible to identify possible candidate circuitries in the brainstem, if any, that may account for such gradient. It is also possible that isoform regulation could be independent of the circuit and each neuron may regulate tau isoforms independently of each other, which needs fundamental reconsiderations in the future studies. Although preceding studies have suggested that the earliest neurofibrillary lesions are detected in the brainstem [6, 19, 46], it is difficult to directly compare the extent of 3R tau dominance between different anatomical structures, because the densities and characteristics of the underlying neuronal population are quite different. Therefore, we focused on the trend of increase for the proportion of RD3 at the coordinated horizontal anatomical levels between cases, rather than direct comparison between the data of the midbrain and pons or between the data of the brainstem and the hippocampus (Additional file 5: Figure S4), and did not attempt to show that brainstem tau lesions develop prior to hippocampus in this study. Rather, we demonstrated that the progressive accumulation of 3R tau over 4R tau is shared between brainstem and the hippocampus.

Biochemical evaluation for 3R and 4R tau by immunoblotting may be potentially interesting and helpful in the future studies. However, considering the small number of NFTs in the AD brainstem, such experiment may require significant numbers and sufficient volume of precious frozen samples from the brainstem (at least midbrain and pontine tegmentum), which are not readily available.

What are the possible mechanisms that regulate isoform-dependent deposition of tau in each neuron? If progressive increase of 3R tau in Alzheimer-type NFT mirrors relative abundance of 3R tau molecule or of its production, it is expected that the amount of mRNA is progressively upregulated in relevant areas. Although data on the brainstem have not been reported, mRNA of 3R tau is not upregulated in AD brains and in normal controls [8, 15, 26, 36, 58]. This raises a possibility that some post-translational modifications of 4R and 3R tau and their interactions may be responsible for such progressive representation of 3R tau over 4R tau. For example, in vitro assembly of 4R tau is significantly decreased by coincubation with 3R tau [1]. Another in vitro study has demonstrated that 4R and 3R tau monomers can grow onto the 4R/3R tau filaments and/or 3R tau filaments, while only 4R tau monomers can grow onto 4R tau filaments [12]. Do the aggregates composed of 4R and 3R tau exhibit corresponding immunoreactivity to both 4R and 3R tau? If an excess of one isoform over the other in the aggregates results in conformational alteration of tau molecules, epitope representation of each isoform could be skewed as if one of the two were absent, even if it is still integrated in the aggregates. Moreover, recent studies on mass spectrometric analyses of sarkosyl-insoluble and protease-resistant tau filaments from human brain with AD and cryo-electron microscopic analyses demonstrated that the protease-resistant cores of tau filaments contained abundant residues of the exon 11–12, which is proposed to adopt a combined cross- β/ β-helix structure with two identical protofilaments, as well as over a dozen of residues on the C-terminal side of exon 10 (4R tau) and exon 9 (3R tau), which exhibit unsharpened density on cryo-electron microscopy like a less ordered β-sheet [14, 48]. These results were compatible with the peptides identified in earlier studies, although there were some variations in the residue length [30, 57]. Still, it is unknown whether these C-terminal fragmentation of tau filaments occurs physiologically to the neurofibrillary changes in vivo. Even if the protease-resistant cores of tau filament are fundamental structures for neurofibrillary changes, how they are related to tau isoforms remains to be clarified because epitopes specific to tau isoforms were not included in the protofilament, a principal constituent of the protease-resistant cores. Because representation of each isoform is modified by aggregation, conformation and fragmentation, which epitopes of tau isoforms are involved in different tau filaments in the human brains may be crucial.

By taking advantage of the dual visibility of quantum dot (fluorescent nanocrystal identifiable also on electron microscopy) used as immunolabeling, we developed a method to correlate fluorescent microscopic images with their exact counterpart on immunoelectron microscopy [33, 49, 55]. This approach will clarify how either 4R or 3R tau epitope is involved in the formation of tau filaments at different stages of their formation in vivo. As double-immunofluorolabeling for 4R and 3R tau successfully demonstrated relationship between the isoforms at light microscopy level, we are now trying to improve this immunoelectron microscopy so that two different epitopes for 4R and 3R tau are visualized in relation to tau filaments of different morphologies and stages. It is expected that in vivo relation between 4R and 3R tau will be better understood if this double-labeling immunoelectron microscopy is successful for tau filaments.

In the present study, the neurofibrillary changes were present without nearby Aβ in 43% of the cases (Table 1). Such discrepancy between tau and Aβ deposition is attracting an increasing attention [3, 7, 9, 13]. Regional distribution of tau [41] and that of Aβ [27, 51] in the brainstem have also been described. Because these had been described separately, it was not clear how tau and Aβ deposits are spatially related. The present study is the first that compared distribution of tau and Aβ deposits directly in neighboring sections of the same cases at different stages of disease progression (Figs. 2, 5, 6). The neurofibrillary changes were frequent in the ventral half of the PAG (Fig. 5a, b, open arrows), while the Aβ deposition was frequent in its dorsal half (Fig. 6l, solid arrow). Although there was an increase in the proportion of 3R tau-positive midbrain neurofibrillary changes with advancing neuritic plaque score (Fig. 3a h, n, b h), Aβ deposition was not directly linked to this increase of neurofibrillary changes, because their distributions were discrepant in the brainstem (Figs. 2, 5, 6). The most striking example of such discrepancy was detected in a case of centenarian (102 years old, case 21, Table 1), which showed a clinical history of dementia, 3R tau-dominant brainstem NFTs, NFT stages V/VI, but neuritic plaque score 0 and the absence of Aβ deposition at the midbrain and pons (Fig. 6k). Taken together, our observations suggested that the tau deposition and the progressive dominance of 3R tau-positive lesions may occur independently of Aβ deposition in the human brainstem at least to some extent.