Three-dimensional study of Alzheimer's disease hallmarks using the iDISCO clearing method

Detecting three-dimensional beta amyloid plaques in a full mouse brain hemisphere, with anti-Abeta peptide antibodies or Congo red

Here we explored the possibilities of studying amyloidosis in 3D. We propose that studies of beta amyloid plaques and their surroundings could gain substantially by resolving beta amyloid plaques in their three-dimensional context. We applied iDISCO (immunolabeling-enabled three-dimensional imaging of solvent cleared organs), a method developed (Renier at al., 2014) to combine whole-mount immunolabeling with the high transparency offered by the solvent-based clearing technique 3DISCO (Erturk et al, 2012). We first tested eleven anti-Abeta antibodies for compatibility with the iDISCO methanol-based immunostaining protocol for their capacity to reveal beta amyloid plaques in large mouse tissue blocks (see Table S1). Based on label specificity and effective tissue penetration (see Figure S1), one antibody was selected (referred to as PAb #10) for Abeta labeling throughout the study. Full mouse hemispheres were then prepared and treated with PAb #10 and imaged utilizing high-speed light-sheet microscopy, resulting in image data sets that enable optical isolation of arbitrary image planes (optical slice) or volumes throughout the brain. A representative sagittal optical slice obtained from a 27-month-old 2xTg mouse brain reveals the presence of beta amyloid plaques in the cortex and hippocampus (Figure S2A, top panel). A magnified view of this optical slice shows a large concentration of amyloid plaques throughout the cortex and in the hippocampus (Figure 1A, top image). A direct comparison (two hemispheres from the same mouse brain) between the most promising antibodies (PAb #10 and PAb #9) showed a similar distribution even though many fewer amyloid plaques are visible in the center of the sample labeled with PAb #9, presumably due to reduced penetration depth for PAb #9 as a result of antibody depletion (Figure S2B, S3).

Next, in an attempt to circumvent some of the limitations inherent to antibodies (cost, slow diffusion and occasional unspecific surface labeling), we tested the possibility that a small molecule known to bind beta amyloid plaques, Congo red, could be effectively used in combination with iDISCO, as demonstrated with other small molecules and clearing protocols (Jahrling et al, 2015). We performed a comparison between Congo red and PAb #10 on the two hemispheres of a 27-month-old 2xTg brain. Single optical slices and orthogonal planes reveal that both antibody and Congo red stainings of amyloid plaques are visible throughout the volume of the hemispheres (Figure S2A). Magnified views of cortices and hippocampi clearly demonstrate the specificity of both antibody and dye staining of plaques (Figure 1A) when compared with control wild type tissue (Figure 1B). Maximum projection images highlight the distribution of staining in the intact large tissues. Staining of full hemispheres with PAb #10 and Congo red is shown in an aged 2xTg AD mouse and an aged wild type control mouse (Figure 1C). Highlighting the time efficiency of using Congo red for plaque visualization, we found effective staining throughout an intact hemisphere after only two hours of dye incubation (Figure 1D).

To illustrate the feasibility of increased lateral resolution imaging, whole hemispheres were also imaged using confocal microscopy. Optical slices further confirm the specificity of PAb#10 labeling in the 2xTg AD mouse brain compared to wild type brain staining (Figure S3D). Maximum intensity projections and animated volume rendering of intact brain hemispheres reveal plaques with exceptional detail, combining the resolving capabilities of confocal microscopy with the volume capacity enabled by clearing (Figure S3E-F and Movie S1), though axial resolution is reduced and acquisition times are longer compared to light sheet microscopy.

To directly compare the efficacy of the antibody versus small molecule dye staining of beta amyloid plaques, we simultaneously applied both labels to a single brain hemisphere of a 2xTg AD mouse. Optical slices and volume projections within the hippocampus and cortex demonstrate a comparable efficacy of the two labeling strategies for targeting large amyloid plaques deep in the brain (Figure 1E-F and Movie S2). Based on these results, the two staining methods are generally comparable. Interestingly, detailed inspection of enlarged images of a cortical region reveals differences in labeling patterns between the two methods. Magnification of thin optical slices shows prominent labeling of the plaque core in both cases, whereas subtle outer structures surrounding the core are seen only with antibody staining, suggesting a greater sensitivity of the antibody detection method (Figure 1G). Surface rendering further emphasizes the 3D differences between the two stains and clarifies the plaque ramifications (Figure 1H; cyan) visible around the core of the plaque (Figure 1H; red) and surrounding beta amyloid deposits labeled by antibodies (Figure 1H; green).

Next, as immunohistochemistry (IHC) is the prevailing method of choice to detect beta amyloid plaques in brain tissue, we compared quantitation of plaques in physical sectioning and DAB staining versus optical sectioning in age-matched animals. In one 17-month-old 2xTg AD mouse, we used iDISCO to stain plaques with both Congo red and PAb #10 and acquire serial sagittal sections (Figure 2A). An additional animal was treated according to standard IHC sectioning and staining for plaques (Figure 2B). A comparative quantification of antibody-labeled plaques in tissue sections versus matched optical planes from iDISCO labeling validates that there is statistically equivalent plaque detection between strategies (Figure 2C). When plaque staining is combined with the autofluorescence in the green spectrum, iDISCO provides optical sectioning with detailed plaque morphology as well as local topography. Volume imaging of plaques with iDISCO also accommodates additional labels, such as nuclear staining to reveal neighboring cell density (Figure S4A).

As cleared samples are imaged intact (without sectioning), a more complete and accurate morphological characterization of beta amyloid deposits is possible. iDISCO permits staining and imaging of large intact tissue volumes (Figure 2D), upwards of one cm³, while simultaneously maintaining resolved details of individual amyloid deposits. Furthermore, representative maximum intensity projections and surface renderings of individual stained structures from small image volumes reveal the diversity of amyloid deposits found in the brain, ranging from plaques of varying sizes and shapes, integrated or not in larger networks (e.g. vasculature deposits) (Figure 2E and Movie S2). iDISCO is compatible with a multitude of accessible microscopy techniques, as demonstrated with laser scanning multiphoton (Figure S4B) and confocal (Figure S4C) microscopies. With each imaging system, we can access volume data, while resolving a variety of plaque sizes, morphologies and aggregates, as well as other relevant parameters such as vascular deposits.

Evaluating dynamics and regionality of plaque formation and distribution in the whole mouse brain

We evaluated the development of beta amyloid plaques in the 2xTg AD mouse line starting with a time at which mice are usually considered beta amyloid plaque-free and ending with old mice with high amyloid plaque burden. Maximum intensity projections of the entire hemispheres show a gradual progression of amyloid plaque staining at each time point (Figure 3A, left panel). With whole hemisphere image data, smaller volumes of interest can be digitally segmented to compare the plaque development profile in any of the prominent structures of the brain (Figure 3A). With images of the complete population of beta amyloid plaques in the intact brain hemisphere, we can directly quantify plaque abundances. The object detection algorithm from the volume image analysis software (Imaris spot detection) was applied to each of the digitally dissected regions of our AD time profile image dataset (Figure 3B). Quantification of each region and age shows the time course of amyloid plaques appearance in the 2xTg mouse model, starting in the cortex and, surprisingly, increasing in the cerebellum earlier than in the hippocampus (Figure 3C). Furthermore, with volume analysis we have the ability to evaluate distances, in 3D space, between plaques in a given region over time (Figure 3D).

Automated whole brain quantitation using ClearMap

Whole brain image data sets require manual input for region-specific quantitation. To circumvent this, we applied a recent tool to streamline region segmentation and allow automated mapping of segmented data, such as amyloid deposits, to annotated references such as the mouse Allen Brain Atlas (Renier et al., 2016). This available toolbox for python, called ClearMap, uses open source modules to rapidly align the autofluorence reference channel from the aligns light-sheet stacks to a reference brain (here, a STP reference aligned to the Allen Brain Standard Annotation). To demonstrate the applicability of ClearMap for comparative studies on plaque distribution, we performed a time course evaluation of 2xTg AD mouse brains using rapid Congo red staining. Intact hemispheres from mice of increasing age were incubated overnight for convenience (shorter incubation times are sufficient) in Congo red before clearing and light-sheet imaging (Figure 3E). In addition to atlas alignment, ClearMap also provides object detection, which works especially well for regular-shaped objects like spherical cell bodies or beta amyloid plaques. Quantitation can then be reported for any region of interest annotated in the Allen Brain Atlas. In this case we present the twelve regions containing the largest plaque burden in the oldest mouse tissue (Figure 3F). The combination of Congo red staining, volume imaging and ClearMap analysis should help to streamline the analysis of the plaque burden in mouse models of AD, as a brain can be prepared overnight, scanned in under two hours on the light sheet microscope, and analyzed in less than one hour.

Combined volume imaging of beta amyloid and microglia detection

High resolution confocal imaging of combined beta amyloid and To-Pro-3 staining revealed that each plaque is surrounded by a perimeter of nuclei forming a nest-like ring (Figure 4A, circles), suggesting recruitment or displacement of cells, possibly microglia, by the plaques upon amyloid buildup. Though activation of microglia has been associated with AD since the mid 80s (Vostrikov 1985; Rozemuller et al., 1986), the role of reactive gliosis in beneficial plaque clearance or adverse neurotoxicity is yet to be fully clarified. Staining for glia in wild type tissue revealed a homogeneous dispersion of densely labeled cells with discrete arborization (Figure 4B, left image). In comparison, microglia the 2xTg model form clustered patterns of more diffusely labeled cells with enlarged bodies and less discrete branching (Figure 4B, right panel). Upon comparison with the distribution of beta amyloid, it is evident that the clustered microglia are principally associated with amyloid plaques. High resolution volume imaging and surface rendering emphasize that microglia surrounding the plaques enlarge and envelop the plaques, in contrast to the smaller, isolated microglia distant from amyloid plaques (Figure 4C and Movie S3).

Combined volume imaging of beta amyloid and axonal dystrophy

As mentioned previously, it is crucial to re-consider the way we analyze beta amyloid plaques while emphasizing their surroundings. Among the numerous mechanisms contributing to the dementia associated with Alzheimer's disease is the deterioration of axon projections in proximity of plaques described as dystrophic neurites (For a review see Reitz, et al., 2014). Visual access to axons in intact tissue will aid studies looking to characterize the toxicity of plaque buildup or evaluate strategies for reversing axonal dystrophy (Brendza et al., 2005). Simultaneous imaging of beta amyloid plaques and axon filaments was demonstrated with Congo red and neurofilament H antibodies, respectively (Figure 4D). Owing to the high density of axons and abundance of neurofilament antibody binding sites, labeling is dramatically reduced beyond 1 mm from the tissue surface. Completeness of filament labeling is enhanced by sectioning tissue into blocks. Filaments detected in regions void of beta amyloid plaques were regular in diameter and largely undisrupted (Figure 4E), whereas filaments surrounding prominent plaques were markedly dystrophic (Figure 4F).

Evaluation of amyloid plaques and the complete brain vascular network

Because brain vasculature and associated abnormalities remain an important theme in the deposition and clearance of beta amyloid, we included a strategy for labeling the full network of brain vasculature. Intact adult 2xTg brains were labeled for amyloid plaques and blood vessels, providing complete access to the entire brain vasculature network and total plaque deposition. Combined plaque and vasculature imaging was demonstrated with beta amyloid antibodies and Congo red (Figure 4G and Figure 4H, respectively).

Optimal tissue evaluation would maximize the information extracted from single brain volumes. A few studies have suggested the removal of antibodies for serial labeling of multiple targets (Chung et al., 2013a; Murray et al., 2015), though this approach could potentially lead to incomplete stain removal and difficult image interpretation, especially in intact mouse brains. Previously used as a background image for topographical reference, the green spectrum channel can also be utilized for a third label. Here we present simultaneous labeling of beta amyloid plaques, microglia and vasculature for effective triple-labeling of the entire brain hemisphere. Maximum intensity projections from sagittal optical sections, rendered columns, and animations within the mouse brain hemisphere show efficient labeling of all three stains throughout the tissue (Figure 5A-D and Movie S4). We observed a large occurrence of co-segregation between the three parameters. Samples of magnified optical sections from the hippocampus show clustered reactive microglia and blood vessels coincident with amyloid plaques (Figure 5A, 5E upper panel). Surface rendering of a magnified optical volume captures the near complete encompassing of plaques by the recruited microglia and the immediate proximity of nearby vessels (Figure 5E lower panel and Movie S4). The flexibility of iDISCO for evaluating AD mouse models allows for comparison of a large list of potential targets. As a demonstration of an alternate combination for full brain analysis, we stained an intact 2xTg AD mouse hemisphere for vasculature, cell nuclei and beta amyloid (antibody PAb #10). From a single block of imaged tissue, we can evaluate amyloid plaque correlation with the intricate vascular network and regional cell density (Figure S5A-B). Importantly, this can be achieved from the scale of the whole brain down to the resolution of a single cell (Figure S5A, S5C).

Variation of three-dimensional amyloid patterns (TAPs) in archived brain tissues from human AD patients

Clearing methods have been previously applied to mouse embryos and various adult mouse tissues including brains. Gaining insight into amyloidosis in AD pathology also necessitates the evaluation of human tissues. With a collection of frozen post-mortem tissue blocks from nine patients, we applied iDISCO for 3D fluorescence analysis of amyloid plaques in large volumes of human brain tissue. All archived AD human samples at our disposal were of hippocampal origin. It is critical to note that these were fresh-frozen samples, none of which had been handled or prepared in any specific way prior to iDISCO application. Most samples were effectively labeled, cleared and imaged with the iDISCO method. Notably, interesting features of amyloid staining with PAb #10 were revealed; perhaps most striking was the presence of complex tertiary structures forming three-dimensional amyloid patterns (TAPs) of up to 27 mm³ (>Figure 6A). Volume imaging also revealed a remarkable 3D diversity of individual amyloid plaque morphology and 3D formations ranging from dense, strongly labeled patches to less intensely labeled and more isolated plaques. 3D images reveal TAPs that appear as dense volumes, thin sheets, parallel ribbons, and density gradients (Figure 6A, see patient 3, 8 and 9, respectively). High magnification volume projection images again reveal pronounced differences in beta amyloid plaque morphology and size. As with the AD mouse model, staining appeared as large and small spheres, thin filament-like structures and vasculature deposits (Figure 6B). Surface rendering highlights the dramatic differences in density, shape and size between samples (Figure 6B and Movie S5). In addition to visualization of beta amyloid plaques in large volumes of human tissue, volume imaging enables direct measurement of amyloidosis parameters such as individual plaque volume, inter-plaque distance and morphological characteristics like sphericity (Figure 6C). Inclusion of additional markers can increase the informative value of amyloid volume imaging. As with the mouse tissue, simultaneous imaging of multiple stains in human tissue can provide anatomical reference or enable correlative evaluation. As proof of principle, we present successful double staining of a human tissue block with PAb #10 and vasculature (Figure 6D). Volume imaging of amyloid plaques in human brain tissue can feasibly be combined with any of the previously demonstrated markers or other iDISCO-compatible stains of interest.

The origin of AD is still largely debated and scientists remain divided between two main hypotheses (Aβ oligomers and amyloid plaques versus Tau-dependent neurofibrillary tangles), while other important parameters such as inflammation and hormonal regulation among others are also important. Accordingly, we also applied iDISCO for volume evaluation of tau in AD tissue, starting with validation of anti-tau antibody compatibility in 3xTg AD mice (see Table S1). Optical slices from cleared brain hemispheres (29 months old mice) obtained after labeling with anti-tau antibodies show prominent staining in pyramidal neurons in CA1 of the hippocampus as well as positive staining in the neighboring subiculum and the stratum oriens (Figure 7A and Movie S6). Full hemisphere imaging shows the expected distribution throughout the hippocampus and around the entorhinal cortex (Figure 7B). Double labeling of 3xTg mouse brains (29 months old) with Congo red and Tau antibodies shows a highly restrictive parallel distribution of beta amyloid plaques and tau positive cells in the subiculum (Figure 7C, D and Movie S6). Volume images reveal that the pattern of plaque distribution is dramatically influenced by the selected animal model, as seen when comparing 1xTg, 2xTg and 3xTg AD mouse models (Figure S6). Interestingly, co-labeling for the two AD hallmarks in human AD brain samples reveals a degree of regional co-localization. Tau staining is clearly prominent in dendritic branching of individual cells as well as thin axons projecting through the cleared tissue blocks, possibly highlighting the presence of neuropil threads. (Figure 7E).

Three-dimensional perspective on neurodegeneration

We have combined the iDISCO clearing method, light-sheet microscopy, and biochemical and chemical tools relevant to AD (e.g. beta amyloid antibodies, tau antibodies, Congo red, and axon, vasculature and microglia detection) to examine the diseased brains of several AD mouse models and patients in three-dimensions. This revealed clear differences between mouse brains from AD models and material from patient brains. Our volume image analysis unveiled strikingly large patterns of amyloid staining in human brain samples, which we refer to as Three-dimensional Amyloid Patterns or TAPs. We did not observe comparable large amyloid patterns in any of the mouse AD models. This inconsistency highlights a possible difference in the aggregation process between human and mouse species. However, it is not clear if the time required for the TAPs to form is compatible with studies carried out on mice due to their short life expectancy.

AD mouse models present clear differences compared to human samples

AD mouse models are in constant need of refinement to address the biochemical, temporal and phenotypic discrepancies that exist between those mouse models and AD pathogenesis (for a review see Ashe et al., 2010). Beta amyloid plaque distribution can also dramatically differ between the animal models (Fig S6). The present work reinforces the notion that better mouse models are needed, and it offers the possibility of efficiently and globally evaluating the anatomical and temporal distributions of AD hallmarks in promising mouse models (Liu et al., 2015) and future genetic models aiming to better recapitulate the pathology presented in humans.

Amyloid plaques, microglia and vasculature

The immune surveillance provided by microglia is an important topic for AD pathology (for a review see Aguzzi et al., 2013). Here we present detailed 3D renderings and videos that unequivocally demonstrate strong association between microglia and beta amyloid plaques. It is remarkable to observe the targeted activation of microglia in contact with plaques. The precise role(s) of microglia segregating around plaques remains unclear, but their visualization in 3D opens avenues to better understand where, and perhaps, how they are recruited.

Cerebral amyloid angiopathy (CAA) is an integral component of a series of important disorders strongly associated with dementia as well as cerebral hemorrhage. It is known that, in the case of AD, some arteries are more targeted by deposition of beta amyloid plaques than others (Viswanathan et al., 2011; Yarchoan et al., 2012). We demonstrated that we could visualize such associations and that this is facilitated by 3D volume rendering. As such, the present study provides a relevant approach to examine the relationship between amyloidosis and CAA. Additionally, further 3D characterization will help clarify why the amyloid plaques congregate around blood vessels. Do they have a specific affinity for blood vessels or, on the contrary, do the plaques appear as a consequence of other phenomena such as local inflammation, micro-hemorrhage, etc?

Tau and neurofibrillary tangles

In addition to the effort dedicated to the Abeta peptide amyloidosis in this study, we also examined Tau staining in 3D. While we do not aim here to resolve the long standing Abeta-tau debate, tau and neurofibrillary tangles are important in disease progression, if not generation, and therefore warrant examination. It is obvious that the multiple phosphorylation states of Tau considerably increase the complexity of the problem, especially regarding the antibody testing for our current approach (Jicha et al., 1997; Espinoza et al., 2008). We have, however, demonstrated that iDISCO is an effective method for labeling tau with a number of total and phospho-specific tau antibodies, allowing for more detailed evaluation of the volumetric formation of neurofibrillary tangles in the context of AD mouse models and human tissues, a necessary contribution toward complete understanding of the disease.

Future studies

Future studies analyzing large numbers of samples anatomically identical and from various origins will be necessary to fully characterize the extent of these TAPs in human samples. Such large-scale evaluation might allow correlation studies, such as indexing TAPs and ultimately matching them with specific symptoms or clinical traits documented prior to death. While diagnosis of individual patient TAPs might remain limited mostly to post-mortem samples and brain biopsies, we do foresee that comprehensive retrospective analyses may even help categorize patients based on specific three-dimensional pathological features, will contribute to the understanding of some clinical trial failures by retrospective identification of responsive or non-responsive groups, and, ideally, might aid the categorization of patients for future clinical trials. It will be also informative to compare familial versus sporadic AD TAPs, and determine if specific familial AD mutations correspond to unique TAPs. Furthermore, a similar approach using iDISCO clearing should be easily transposable to any form of amyloidosis or related diseases, including non-AD dementias such as vascular dementia (O'Brien et al., 2015), frontotemporal dementia (Bang et al., 2015), dementia with Lewy bodies (Goldberg et al., 2015) and transmissible dementia (Watts et al., 2014; Sanders et al., 2016). This work also may have the potential to advance the evaluation strategies for therapeutic approaches in pre-clinical stages such as immunotherapies (Wisniewski et al., 2015).

In conclusion, we believe that the discovery of large TAPs and the surprising complexity of amyloid plaques in human (but not mouse) brain samples may help revisit disease classification and lead to a better understanding of the initial causes of AD. In more general terms, the methodology developed here to detect beta amyloid plaques in their three-dimensional context, both in murine and human brains, and allowing for the study of up to three markers at a time, including tau protein, is a valuable approach to tackling various characteristics of Abeta amyloidogenesis, and amyloidogenesis in general.

Mouse line genotypes and related work

All procedures were approved by the Rockefeller University Institutional Animal Care and Use Committee. All mice were housed with a 12-hour light/12-hour dark cycle and fed ad libitum. Transgenic mouse lines (2xTg: APPswe/PSEN1dE9; 1xTg: Tg2576 harboring HuAPP695.KM670/671NL; 3xTg: PS1M146V, APPswe and tauP301L], and C57BL/6J WT mice were obtained from The Jackson Laboratory (Bar Harbor, ME, USA). Mice were anesthetized via Nembutal IP injection (50-100 µl) and intracardiac perfusion was performed first with 20 ml ice cold PBS then 20 ml ice cold PBS with 4% paraformaldehyde. Dissected cerebral tissue was post-fixed overnight in PBS with 4% PFA at 4°C followed by 1 hour at room temperature.

Human tissue origin and related work

Frozen human hippocampal brains blocks were obtained from the New York Brain Bank (Columbia University, Alzheimer Disease Research Center, Taub institute) in agreement with our Institutional Review Board. The brain tissues were collected from autopsied patients with AD clinical diagnosis as well as post-mortem histo-pathological signature for AD. Frozen tissue kept at −80°C was cut into blocks approximately 0.125 cm³ and defrosted on ice for 20 minutes before fixation in PBS with 4% PFA overnight at 4°C and 1 hour at room temperature.

Tissue permeabilization

Tissue was pretreated following the methanol protocol provided in Renier et al. (2014). Briefly, whole brains were dehydrated with gradual addition of methanol in PBS (50% x1, 80% x1, 100% x2, each for 1 hr). Overnight bleaching with a 1:5 ratio of hydrogen peroxide:methanol was done at 4 °C. Tissue was the n gradually rehydrated in PBS by removing methanol at 20% increments (1 hr for each step, followed by 2 additional washes in PBS). Detergent washing was then performed in PBS with 0.2% Triton X-100 (2× 1 hr).

Immunolabeling

To block any residual active aldehyde groups, tissue was incubated overnight at 37°C in PBS with 0.2% Triton X-100 and 0.3 M glycine, followed by blocking in PBS with 0.2% Triton X-100 and 6% normal goat serum. Blocking and staining incubation times were 7 days for whole hemispheres and 2 days/mm (maximum of 7 days) for sectioned blocks. For 7 day incubation steps, 0.02% sodium azide was included to prevent contaminants. Following blocking, the tissue was washed for 1 hr twice in PBS with 0.2% Tween-20 and 10 μg/ml heparin (PTwH). Antibodies were applied for the indicated time in PTwH with 3% NGS at 37°C with rocking. Excess primary antibodies were washed for 1 day in PTwH with periodic solution changes before starting incubation of secondary antibodies. After secondary antibody labeling, tissue was washed in PTwH for 2 days with periodic solution changes.

For vasculature staining, tissue was fixed in 4% PFA without perfusion to preserve the endogenous immunoglobulins. Dye-conjugated secondary antibodies (1:100 dilution) against the tissue species were then used without need of primary antibodies. There were no detectable differences in amyloid plaque and glial staining in the non-perfused tissue.

Primary antibodies against amyloid plaques and neurofibrillary tangles were selected by evaluation on 1-2 mm tissue sections after methanol pretreatment described in the iDISCO protocol (Renier et al., 2014) and are listed in Table S1. Microglia and neuronal axons were labeled with rabbit Iba1 antibodies from Wako Chemicals and chicken neurofilament antibodies from EMD Millipore. Secondary antibodies were selected based on the number of labels in a given sample. Alexa 647, if available as a conjugate against the respective primary antibody, was prioritized for single labels, followed by Alexa 568 as a second label, and Alexa 488 as a third label as demonstrated with vasculature staining. Autofluorescence was imaged using 488 nm excitation, eliminating the option for simultaneous Alexa 488 staining. Secondary antibody concentrations were matched to that of the respective primary antibody, with a maximum limit of 20 μg/ml.

For samples additionally stained with chemical dyes, the respective dye was included in the secondary antibody incubation step. For samples stained solely with a chemical dye, the respective dye was incubated at 37°C in PT wH for 2 days before 2 days of washing and clearing. On samples labeled with dye for volume imaging, Congo red and To-PRO-3 were used at 10 μM and 2 μM, respectively.

Standard immunofluorescence labeling was done on 45 μm cryosections. Sections were washed in PBS for 20 min and PBS with 0.2% Triton X-100 for 20 min then blocked in PBS with 6% normal goat serum and 0.2% Triton X-100 for 1 hr, all at RT with rocking. Beta amyloid was stained overnight at 4°C in PBS with 3% normal goat serum and 0.2% Triton X-100 (anti-beta amyloid, 1:250; Congo red, 3.3 μM). Slices were washed 3 x 10 min in PBS prior to secondary labeling (goat anti-rabbit Alexa 647, 1:250) for 2 hr at RT and washed in PBS 3 × 10 min before imaging.

Immunohistochemistry

Mouse brains were fixed in 4% paraformaldehyde (Electron Microscopy Sciences, Hatfield, PA, USA), washed with PBS, and incubated in 25% sucrose for 48 hours. Brains were then frozen on dry ice with freezing medium (VWR, Radnor, PA) and sliced in 50 μm sections. Sections were stained using PAb #10 and a DAB peroxidase substrate kit (Vector Laboratories). Sections were imaged using a wide field microscope at 2.5x with a QI Click color digital camera. Images were assembled using the Fiji plugin for ImageJ (Preibisch et al., 2009).

Tissue clearing

After washing away excess secondary antibodies, optical clearing of iDISCO samples was performed as described in Renier et al. 2014. Tissue was gradually dehydrated in resistant glassware with tetrahydrofuran in water (50% x1, 80% x1, 100% x2, 1 hour each). Remaining lipids were extracted with dichloromethane for 1 hour, and dibenzyl ether (DBE) was used for refractive index matching. Samples were kept in a full vial of DBE. Samples stained with Congo red for automated ClearMap analysis were treated with a modified protocol to allow for anatomical alignment described in Renier at al. 2016. The most significant change is removal of tetrahydrofuran to avoid shrinkage of some tissue layers that prevents effective atlas matching.

Imaging and image processing

Images of large samples were principally acquired on a light-sheet ultramicroscope (Ultramicroscope II, LaVision Biotec). High resolution images were takes on a confocal (Zeiss 710 LSM) or multi-photon (FV1000MPE Fluoview, Olympus). The light-sheet and multi-photon microscopes were provided by the Bio-Imaging Recourse Center of The Rockefeller University. All cleared tissue images were acquired with tissue submerged in DBE. Volume images were acquired with 16-bit depth to allow for a broad range of intensity values and rendered using Imaris software from Bitplane. The ultramicroscope was used with a fixed 2x/0.5 N.A. MVPLAPO objective. Exposure time was 100 ms for each frame, spaced axially with a 2.5 μm interval. Laser intensities were set to prevent saturated pixels (typically 50 to 70% of maximum power). The following laser excitations were used: autofluorescence and Alexa 488, 488 nm; Alexa 568, 561 nm; Alexa 647, 640 nm. Images were taken with an Andor NeosCMOS (2160 × 2560 pixels).

For visualization of volume images, gamma corrections and background subtractions were applied to reveal the range of staining intensities and remove global background signal, respectively, for each data set. Automated amyloid plaque detection was done with the Imaris ‘spots’ feature, using an estimated diameter of 30 μm and minimum threshold values for maximum peak intensity (3,000) and standard deviation (500) that most accurately detected visible plaques in random test regions without including background signal. Digital dissection of volumes was done with Imaris by manual tracing of autofluorescence channels at appropriately spaced planes, typically 250 μm and 50 μm for large and small surfaces, respectively. Surface rendering for small volumes of amyloid plaques, glia and vessels was automated on Imaris, with manual threshold adjustments for each channel to remove background.

Automated quantitation with ClearMap

ClearMap, a light-sheet image analysis toolbox, is available upon request by authors (Renier et al., 2016). The free package operates on a Linux operating system with python programming software. Light-sheet images were collected from autofluorescence (488 nm at 100%, 50 ms exposure) and Congo red (561 nm at 100%, 50 ms exposure, this time with a 20-point horizontal focus and contrast blending). Autofluorescence images were applied for automated alignment to the Allen Brain Atlas, and Congo red images were used for object detection. ClearMap detection parameters for the plaques were: background subtraction (7 pixels), difference of Gaussian filter (4,4,6), extended maxima h-max (20) and size (5 pixels), cell shape threshold (125).