X-ray multiscale 3D neuroimaging to quantify cellular aging and neurodegeneration postmortem in a model of Alzheimer’s disease

Purpose
 Modern neuroimaging lacks the tools necessary for whole-brain, anatomically dense neuronal damage screening. An ideal approach would include unbiased histopathologic identification of aging and neurodegenerative disease. Methods We report the postmortem application of multiscale X-ray phase-contrast computed tomography (X-PCI-CT) for the label-free and dissection-free organ-level to intracellular-level 3D visualization of distinct single neurons and glia. In deep neuronal populations in the brain of aged wild-type and of 3xTgAD mice (a triply-transgenic model of Alzheimer’s disease), we quantified intracellular hyperdensity, a manifestation of aging or neurodegeneration. Results In 3xTgAD mice, the observed hyperdensity was identified as amyloid-β and hyper-phosphorylated tau protein deposits with calcium and iron involvement, by correlating the X-PCI-CT data to immunohistochemistry, X-ray fluorescence microscopy, high-field MRI, and TEM. As a proof-of-concept, X-PCI-CT was used to analyze hippocampal and cortical brain regions of 3xTgAD mice treated with LY379268, selective agonist of group II metabotropic glutamate receptors (mGlu2/3 receptors). Chronic pharmacologic activation of mGlu2/3 receptors significantly reduced the hyperdensity particle load in the ventral cortical regions of 3xTgAD mice, suggesting a neuroprotective effect with locoregional efficacy. Conclusions This multiscale micro-to-nano 3D imaging method based on X-PCI-CT enabled identification and quantification of cellular and sub-cellular aging and neurodegeneration in deep neuronal and glial cell populations in a transgenic model of Alzheimer’s disease. This approach quantified the localized and intracellular neuroprotective effects of pharmacological activation of mGlu2/3 receptors. Supplementary Information The online version contains supplementary material available at 10.1007/s00259-022-05896-5.

X-PCI-CT data collection: one half brain X-PCI-CT dataset per animal was measured with the 3.0 3 µm 3 voxel setup (for a total of 15 half-organ 3D datasets). In addition, local tomographic scans aiming CTX and HIP layers were performed with the 0.7 3 , 0.3 3 and 0.1 3 µm 3 voxel setups, respectively on 7, 4 and 3 animals. The limited beam-time available at high-resolution synchrotron X-PCI-CT imaging setups was the limiting factor in the collection of the higherresolution data.
Setup descriptions: 1. 3.0 3 and 0.7 3 µm 3 voxel micro-X-PCI-CT data collection: the single-distance propagationbased (PBI) X-ray phase-contrast CT imaging setup of the ID17 Biomedical beamline 6 of the European Synchrotron (ESRF -Grenoble, France) was used to collect full-organ mouse brain X-PCI-CT data with effective voxel size of 3.0 3 µm 3 , as well as subsequent partialorgan maps with the 0.7 3 µm 3 voxel optics system. All 15 half brain samples from aged animals (7 3xTgAD and 8 B6/129 WT mice) were imaged in their entirety with the 3.0 3 µm 3 setup, and 7 of those samples were imaged with the 0.7 3 µm 3 setup, aiming local CT scans to dorsal and ventral CTX and HIP brain regions. 3.0 3 µm 3 voxel micro-X-PCI-CTs were performed in the imaging hutch of the ID17 beamline, around 150 m away from a wiggler X-ray source, using a quasi-parallel, quasi-monochromatic 30 keV X-ray beam, obtained from a Si double Laue crystal monochromator system 7 , and 1 mm Al and 0.8 mm Cu as absorption filters. 0.7 3 µm 3 voxel micro-X-PCI-CTs were instead performed in the so-called MRT hutch of the ID17 beamline, 45 m away from the wiggler, using a pink X-ray beam with peak at 40 keV and a ~20 keV broad spectrum. In this case, movable absorber filters included aluminum (1 mm), copper (0.7 mm) and carbon (1.15 mm). Both setups use a sCMOS PCO.Edge 5.5 (PCO AG, Germany) detector camera and a YAG-based scintillator, coupled to a x1:2 optic system to obtain the 3.0 3 µm 3 effective voxel size (sample-todetector distance set to 180 cm), and to a x1:10 optic system to obtain the 0.7 3  ID17 data CT reconstruction: tomographic reconstructions were performed using the ESRF PyHST2 8 software package via a standard filtered-back projection after application of the singledistance Paganin 9 phase-retrieval algorithm (PyHST2 Paganin-length parameter: 300).
2. 0.3 3 µm 3 micro-X-PCI-CT data collection: the single-distance PBI X-PCI-CT imaging setup of the TOMCAT beamline 10,11 of the Swiss Light Source (SLS, Paul Scherrer Institute, Villigen, Switzerland) was used to collect cellular-resolution local brain-tissue X-PCI-CT data using an optical system affording an effective voxel size of 0.3 3 µm 3 . Four aged brain samples (2 3xTgAD and 2 WT mice) plus the 1 young brain sample (WT mouse), were imaged by aiming local CT scans at dorsal and ventral CTX and HIP layers. Acquisitions were performed with a 5 cm sample-detector distance, using a quasi-parallel quasimonochromatic 21 keV X-ray beam, obtained with a W/Si multilayer monochromator, and 100 µm Al and 10 µm Fe filters. A sCMOS PCO.Edge 5.5 detector camera, coupled to an Optique Peter microscope at 20x magnification and a 20 µm-thick LuAG:Ce scintillator, was used to obtain an effective voxel size of 0.325 3  TOMCAT data CT reconstruction: tomographic reconstructions were performed via filtered-back projection after application of Paganin's single-distance phase-retrieval algorithm, using in-house computing resources available at TOMCAT.
3. 0.1 3 µm 3 nano-X-PCI-CT data collection: the X-ray nanoholotomography (XNH) imaging setup of the ID16A nano-imaging beamline [12][13][14][15] of the ESRF was used to collect local braintissue X-PCI-CT data with effective voxel sizes of 0.1 3 µm 3 . 3 aged brain samples (2 3xTgAD and 1 WT animals), were imaged by aiming local CT scans at dorsal CTX and HIP layers within dissected 2 x 2 x 4 mm 3 -volume tissue biopsies. The ID16A XNH setup 13,14 uses multilayer-coated Kirkpatrick-Baez mirrors to focus the beam to a high-brilliance ~ 30 x 30 nm 2 spot, which enables imaging with sub-100 nm spatial resolution. The beam energy was set to 17 keV, with a monochromaticity of 1%. At a fixed focal-plane-to-detector distance of ~1.2 m, projections images were recorded with a 4kx4k fast readout and low noise CCD camera (FReLoN, ESRF), binned to 2048 x 2048 pixels and coupled to magnifying optics and a 23 µm thick GGG:Eu scintillator. The voxel size, given by the geometrical magnification, was set to 0.1 3 µm 3 and the corresponding field of view was 0.2 mm x 0.2 mm (H x V). Samples were placed in vacuum (~10 -7 mbar) on a rotating stage downstream of the focal plane. Single-distance low-resolution CT overview-scans with a voxel size of ~200 3 µm 3 helped locate the most relevant CTX and HIP regions, within the extended rod-like samples, which were then further imaged at 0.1 3 µm 3 voxel size. 0.1 3 µm 3 voxel CT scan parameters: sets of angular holograms 16 were recorded at 4 different (predefined) propagation distances 12,17,18 (starting with a 40 mm focal-plane-to-sample distance, and then moving the sample progressively closer to the detector) by rotating the sample over 180 degrees. Exposure time for each projection was 250 ms and 2000 projections were recorded for each tomographic sub-scan. Total scan duration was around 4 h.
ID16A data CT reconstruction: projections were normalized, rescaled to the smallest pixel size, and registered via an in-house algorithm based on cross-correlation. Assuming pure phase object samples, a 4-distance contrast-transfer-function-based algorithm for phase-retrieval was applied 18,19 , including Wiener regularization 20 to improve low frequencies, using GNU Octave software. The final 3D datasets were obtained from angular phase-maps through filtered-back projection CT reconstruction, using PyHST2 8 software.

X-PCI-CT data post-processing and analysis:
CT artefacts: CT cupping artefacts related to local-tomography acquisitions were removed by flattening reconstructed CT images via normalization against their Gaussian-blurred version (filter sigma size: 50). CT ring artifacts were in large part removed from reconstructed CT images with an ESRF in-house post-processing tool 21 .
MIPs: maximum intensity projections (MIPs) were computed by summing 20-100 consecutive CT images via the maximum intensity z-projection function in ImageJ 22 . This approach highlights hyper-intense image features and creates a pseudo-volumetric rendering by projecting the brightest voxels within a 3D stack onto a 2D image.
LUT-recoloring of X-PCI-CT data: different look-up tables (LUTs) available in ImageJ were used in the recoloring of X-PCI-CT images (e.g. 6-shades, Orange Hot, Viridis LUTs).
3D renders: all 3D renderings were obtained using the commercial software VG Studio Max 3.2 (Volume Graphics GmbH, Heidelberg, Germany). Volumetric extraction and segmentation of different tissue features (e.g. hyperdense particles, vasculature, normal parenchyma) were performed with the threshold-based gray-value-range voxel selection option or the regiongrowing voxel selection option. The threshold-based tool was used for hyperdense (HD) particle features and hypodense vasculature features. Normal brain cells and plaque-like deposits were instead extracted from inverted-gray-level X-PCI-CT maps via manual selection and the regiongrowing tool. Color-coding was achieved with customized color presets, generally rendering HD particles in white, vasculature in red and normal parenchyma in azure. 3D renderings were then obtained via either the non-transparent 'Isosurface' or the semi-transparent 'X-ray' rendering algorithms.
Automated 3D segmentation of hyperdensity (HD) within X-PCI-CT data: threshold-based segmentations of HD particles were made automatic by using the auto-thresholding tool in ImageJ and by applying the maximum entropy auto-threshold algorithm 23 (MaxEntropy AutoT), which has already been shown to work on X-PCI-CT data for the segmentation of macroscopic amyloid deposits 2 . We computed the MaxEntropy AutoT algorithm slice-by-slice on every 10 th CT image in a volume of interest, and averaged the obtained oscillating threshold values to choose a single threshold level for the entire volume. In this way, differently-sized HD particles could be extracted.
Quantification of hyperdensity (HD) within X-PCI-CT data: a. Macroscopic plaque-like deposits: to obtain a one-off proof-of-principle X-PCI-CT-based Aβ plaque quantification, macroscopic senile plaque-like HD objects (n=27) were segmented-out of X-PCI-CT datasets from one WT and two 3xTgAD mice, using the manual region-growing tool in VG Studio Max. The volume of each extracted plaque-like HD objects, obtained by counting the voxels within each object mask, was used to compute the object's equivalent-sphere diameter. Diameter values from different objects were combined to obtain a distribution and calculate a mean size for these plaque-like HD objects.
b. HD cell-like particles: to obtain a systematic quantification of particle-like HD within HIP and CTX layers in 3-0.1 µm pixel X-PCI-CT data, every volume-of-interest (VoI) was segmented via the automatic threshold-based approach described above, and the automatically-computed threshold value was used as lower limit to compute a volumetric mask of the data. The segmented mask was then analyzed via the '3D Object Counter' ImageJ plugin 24 , which extracts individual unconnected 3D particles and computes their volumes. Just as for macroscopic plaques, the equivalent-sphere diameter was calculated from the volume of each individual HD particle as a way to express object sizes, and all computed diameters were combined to obtain the distribution of particle-like HD 3D objects within the VoI. Objects with diameters smaller than twice the pixel size (i.e. nonresolved objects) were excluded from distributions. Moreover, the few outlier objects with diameters greater than 30 µm were also excluded from distributions, because deemed to arise from non-cell-like objects (possibly image artefacts, or larger anatomical brain features unrelated to cellular neurodegeneration). Three different quantitative parameters were extracted from HD particle size distributions: i.
HD particle 3D load: the proportion of brain tissue within a VoI pertaining to HD particles, computed by summing the volumes of all extracted individual particles and then normalizing with respect to the total volume of the VoI. Load was then expressed as the percentage of the total volume occupied by the HD particles.
Tri-modal HD-particle size distributions from 0.7 3  algorithm 25 . A 2-component GMM was instead fitted to the bimodal distributions obtained from the 3.0 3 µm 3 voxel-size X-PCI-CT data. In this way, two additional HD-particle parameters could be obtained: ii. HD particle populations mean sizes: the 3-component (2-component) GMM fit of the HD-particle diameter distributions returns three (two) Gaussian means (mu parameters), which express the mean sizes of each of the three (two) cell populations.
iii. HD particle population proportions: the 3 (2) mixing proportions returned from the 3 (2)-component GMM fit of HD-particle diameter distributions, which express the proportion of each HD particle cell population within a VoI.
3D virtual tissue sample collection: sub-volumes of interest (sVoI) were collected from each 3D brain X-PCI-CT dataset, in order to sample four specific brain regions, i. 3D Quantification: volumetric HD particle masks were automatically segmented from each of the (90) sampled sVoIs. As discussed above, individual 3D masks (from each sVoI) were analyzed to extract several sample parameters: HD-particle load, mean HD-particle sizes for three different cellular populations, and HD-particle population proportions.
Quantitative group comparisons: Region-based and animal-group-based averaging of the 3 extracted parameters (load, particle sizes, particle proportions) were performed to study potential differences in the quantified levels of HD-based tissue neurodegeneration between different animal types, different treatment types and different brain regions. Group distributions (and group averages) were obtained by combining the individual parameters of all sVoIs pertaining to a specific group. Possible significant differences in parameters between animal groups were tested by one-way ANOVA testing, with Turkey-Kramer's multiple comparison post- Coarse XFM raster-scans, acquired to further perfect scan aiming, were performed with a 400x400 nm 2 step size and 100 ms dwell time. Definitive 2D fine raster-scans instead were performed with a 120 x 120 nm 2

High-field MRI measurements:
Sample preparation: PBS-stored hydrated half brains from one 13 month old WT and one 13 month old 3xTgAD were embedded in 2% agarose gel in PBS and each placed in a 2 ml Eppendorf Safe-Lock tube. After an initial non-contrast-enhanced (non-CE) high-field MRI-imaging session, the agarose gel was removed and the samples were prepared for contrast-enhanced (CE) MRI (following a reported contrast-enhancement approach 32   ultra-microtome (Leica Microsystems), collected on formvar-coated copper grids and counterstained with uranyl acetate 2% and lead citrate 2%. Sections were observed using a Tecnai12 G2 Spirit Bio Twin (ThermoFischer) microscope operating at 120 kV using an Orius SC1000 CCD camera (Gatan).

Histology and immunohistochemistry:
Coronal sections (10 µm) were cut on a microtome and processed for Thioflavin-S staining. 1. The XFM data (Figure 5e) was pre-processed as described in methods section "XFM measurements". Two animals were used in the analysis (one 3xTgAD at 13 months, one WT at 13 months). No outlier values were removed. Data represent mean intracellular elemental content in two groups (normal cells vs. ICHD cells) for elements P, S, Ca and Fe, and are presented in a bar graph, which plots mean 2D elemental density (bars) ± SD (error bars). Sample sizes in the two groups are n=22 cells for normal cells, n=11 cells for ICHD cells. Statistically significant differences were tested by unpaired two-sided twosample Wilcoxon Test, with P-values less than 0.05 considered significant.
2. The quantification workflow for HD particles was performed as described in methods section "X-PCI-CT data post-processing and analysis". Experimental animal groups (13 month old mice) included WT treated with saline (n=4 animals), WT treated with LY379268 (n=4), 3xTgAD treated with saline (n=3), 3xTgAD treated with LY379268 (n=4).
For each selected VoI (90 VoIs in total), HD objects with diameters smaller than twice the pixel size (non-resolved objects) and few outlier HD objects with diameters greater than 30 µm (non cell-like objects) were excluded from distributions. Data in Suppl. Figure 12e represent two example HD particle size distributions from one cortical and one hippocampal tissue volumes, presented in violin plots, which graph the sphere-equivalent diameter of each extracted HD particle within the two example tissue VoIs.  Fig. 13b presents a box plot quantifying the sphere-equivalent diameter of n=27 macroscopic extracted plaque-like HD clusters (plotted are minimum, maximum, median, first and third quartiles). Suppl. Figure 14 expands the data presentation in Figure 6 on the quantification of intracellular HD in the four experimental animal groups, performing the same statistical testing.

SUPPLEMENTARY RESULTS:
Suppl. Fig. 8: The morphological information in Suppl. Fig. 8 suggests a likely difference in the biological nature of ICHD deposits in WT vs. 3xTgAD mice: on one side likely normal age-related accumulations in WT brains, and on the other side neurodegenerative protein accumulation leading to cell-death in the 3xTgAD mice.

Fig. 4:
The double IHC for Aβ and NeuN, observed especially in 13 month old 3xTgAD mouse brains (Fig.   4b), formed distinctive ring-like patterns around neuron nuclei, which match reported IHC work on 3xTgAD mice [33][34][35] . Similarly to other IHC studies on the 3xTgAD mouse [34][35][36][37] , the double IHC for p-tau and NeuN (Fig. 4d), displayed little somatic p-tau involvement (hardly any co-localization of p-tau and NeuN signals), whereas strong p-tau-positivity within neuron dendrites and axons (p-tau IHC signal present in appendages of neuron somas). Suppl. Fig. 11: MRI is known to permit whole-brain mapping, from which cellular-level microstructures can be inferred 40 . MRI signal hypo-intensity is known to correlate within amyloid agglomerates, reportedly due to localized Fe trapped within the protein aggregates, and is used both postmortem and in-vivo 41,42 as a morpho-functional marker for AD-linked neurodegenerative processes in the brain of animal models 43 (including the 3xTgAD mouse 44 ) and in humans 45 .
Differently from X-PCI-CT, TEM involves use of heavy metal salt staining, embedding in Epon and sectioning of brain-tissue in 100-200 nm thick slices. In tissues from aged WT (Suppl. Fig. 11d-e) and aged 3xTgAD animals (Suppl. Fig. 11f), both techniques visualize euchromatin-rich hypodense nuclei with smooth spherical nuclear envelopes and dense nucleoli.

Fig. 6:
Merging results from all sample-volumes, we obtained distributions of each parameter for each experimental group (respectively Fig 6d, 6e, 6f-g). The volumetric load of ICHD-particles was studied both overall between animal groups (Suppl. Fig. 14), as well as regionally (in CTX and HIP layers, Fig. 6f) and sub-regionally (in dorsal and ventral CTX layers, Fig. 6g).
The influence of different image voxel sizes in the dataset of this multiscale analysis was found to, naturally, affect object-density calculations: higher resolving powers allowed the detection of smaller, and thus of more objects per unit-volume. This did not invalidate the calculations of morphological (particle size) and histopathological (3D load) parameters, which afforded similar results at all imaging-system resolutions (Suppl. Fig. 14).

SUPPLEMENTARY DISCUSSION:
Possible artifacts in the X-PCI-CT data: X-PCI-CT scans, especially at the highest spatial resolutions, deposit conspicuous energy locally (from hundreds to thousands of milligrays, depending on the used pixel size), which may lead to sample heating, local dehydration and shrinkage. Moreover, X-PCI-CT data of fixed paraffinembedded samples may suffer the same differential neuropil shrinkage artifacts typical of histological work.

Non-specificity of the ICHD signal in X-PCI-CT maps vs. other techniques:
Image-contrast in most of the phase-retrieved X-PCI-CT datasets presented here, after application of the Paganin phase-retrieval algorithm 9 , is only semi-quantitative. In terms of relative density levels, the ICHD-bearing particles appeared as denser than adjacent normal brain parenchyma, but also as less dense than typical high-Z bio-materials (e.g. bones, iron-rich bloodfilled vessels, or calcifications 46 ), and of a comparable density to that of neuron nucleoli. This signal intensity suggests a likely protein-based composition for the abnormal aggregates along with possible mild metal involvement (e.g. protein-bound Ca or Fe), all intracellular contents expected in aged and neuro-degenerating brains.
X-PCI-CT-measured signal within ICHD neurons is non-specific and could be a reporter of different cellular conditions, both AD-linked and non-AD-linked 47 . This is not dissimilar to the signal aspecificity of other known histological markers of acute cellular neurodegeneration, e.g. 'dark' argyrophilic Gallyas-silver-positive neurons, which aspecifically report cellular neurotoxicity 48,49 , or Fluoro-Jade-positive, which aspecifically report degenerating neurons 50,51 , or even darkstained neurons in toluidine blue histological sections, which aspecifically report likely cellular condensation [52][53][54] . Overall, the ICHD tissue markings in our data on 3xTgAD mice closely resemble tissues stained with a Campbell-Switzer impregnation, a semi-specific dye for AD neuropathology 54 .

Morphology of ICHD vs. biological considerations:
The correlation of morphological observations to known AD micro-biology enables the

Interest of cellular-level AD imaging in animal models:
Cellular-level visualizations of the smallest micrometric intracellular lesions in early stages of AD, even in animal models, are valuable to both studies on mechanism and on pharmacological treatment of AD. As it pertains to disease mechanism, AD animal models have long been used to help characterize early disease phases, even though e.g. mice have a short lifespan without a clear-cut neurodegenerative phenotype. As it pertains to AD-drug discovery, preclinical research on mouse models has highlighted how beneficial effects can arise with early treatment 56 , a realization which has shifted the focus of some new drug tests toward the pre-symptomatic disease phase and on patients at risk for AD (e.g. solanezumab 57 drug-trial programs).

Aβ and p-tau IHC in the 3xTgAD mouse:
The patterns of protein deposition observed in the collected IHC are unsurprising, since the 3xTgAD mouse model is known 58 to present co-localized intracellular p-tau and Aβ toxic fibrils.
Extra-and intracellular Aβ deposits are reported to arise first, starting in CTX layers and then spreading to HIP ones. P-tau fibrillary buildup, instead, was proved successive to the amyloidopathy, accumulating intracellularly, first in HIP and then in CTX, and often co-localizing with the Aβ-positive cells. More generally, tauopathy and amyloidopathy are known to coexist in age-related processes of neurodegeneration in many animal models.

More on the XFM data:
The collected XFM data ( Of note is how nucleoli, regulation centers of ribosomal neurogenesis, neuronal growth and stress response 59 , and recognizable as single dense and rounded intra-nuclear structures, feature P, Ca and Fe elemental hyperdensity in both ICHD-bearing cells and normal cells (Fig. 4). The smaller round-shaped deposits are normal and physiological, since P is a constituent of ribosomal DNA, Ca signals regulate normal gene transcription 60 and Fe reveals itself in the nucleoli of many normal CAsp neurons 61 . The more extended and diffuse intra-nuclear and near-nucleolar P, Ca and Fe areas within ICHD-bearing cells (Fig. 4), instead, seem pathologic/abnormal, likely driven by nuclear or nucleolar dysfunction and leading to neurodegeneration 59 .
It is known that Fe recycling occurs also within lysosomes, which therefore generally contain relatively large amounts of redox-active iron 62 . Studies indicate that the formation of neuronaging-related lipofuscin granula can be due to the oxidative alteration of macromolecules by oxygen-derived free radicals generated in reactions catalyzed by redox-active iron of low molecular weight 1 . Fe thus clearly plays an important role in both aging and neurodegenerative processes, and interestingly appeared as a significantly hyper-abundant metal in our XFM datasets on ICHD-cells in aging and AD mice. Overall, the XFM data suggest that both non-metallic and metallic intracellular elemental hyperdense deposits are contributing to the increase in intracellular electron density measured within ICHD-bearing cells via X-PCI-CTs.
Novel fixation and embedding protocols established for X-ray spectroscopy and microscopy 63,64 may lead to more reliable absolute measurements of metal content within biological tissues in similar future XFM measurements.

More on TEM results:
The observed nuclear membrane infoldings, in literature referred to as 'nucleoplasmic reticula', can occur either in certain types of healthy cells 65 , or in cells suffering from neurodegenerative laminopathy, which leads to aberrant forms of nucleoskeletal invagination, to heterochromatin relaxation and to neuronal death. Interestingly, tau-induced nuclear envelope invagination 66 and lamin dysfunction have been observed in several tauopathy models 67 , including AD models 68,69 .

Role of group II metabotropic glutamate receptors:
The role of group-II metabotropic glutamate receptors (mGlu2 and mGlu3 receptors) in influencing mechanisms of neurodegeneration is debated, since activation of mGlu2/3 receptors by selective ligand agonists (e.g. LY379268 or LY354740) has downstream effects that could lead both to deceleration and to acceleration of neurodegeneration. mGlu3 receptors, for instance, have the potential to trigger neuroprotective effects and possibly decelerate neurodegeneration, since they can mediate the production of neurotrophic factors (e.g. transforming growth factorβ (TGF-β) and glial cell line-derived neurotrophic factor (GDNF) [70][71][72][73][74][75][76] ) and limit the production of Aβ peptide by stimulating α-secretase activity 77 . Instead, selective activation of mGlu2 receptors has been shown to amplify neuron death processes in culture 72,76 and in experimental models of transient global ischemia 78 . Within microglial cells, mGlu2 receptors are known to activate the release of neurotoxic cytokines 79,80 .

ICHD signal intensity and interpretation:
ICHD deposits exhibit levels of X-PCI-CT image gray-level hyper-intensity, at which normal cytoarchitecture is still well visible. This observation implies that the two different density levels are somewhat comparable, and suggests an endogenous near-to-normal biological origin for the ICHD. More extreme levels of hyperdensity, e.g. within metals, would instead likely overpower X-PCI-based image contrast and hide the relatively-hypodense neuroanatomy of a normal nervous tissue. This is not the case, though, for the ICHD observed here. Therefore, an interpretation of the ICHD signal as protein aggregation seems fitting to its signal-level.

P vs. Fe contributions to X-PCI-CT signal:
Quantitatively, a higher absolute value of the mean P intracellular area density (~25 ng/mm 2 ) was observed within ICHD-bearing cells, compared to the value of the mean Fe intracellular area density (~1.4 ng/mm 2 ). This observation may seem to point to a stronger link between X-PCI-CTdetected ICHD and P accumulation, than between ICHD and Fe deposition, and therefore to strengthen an interpretation of the ICHD signal as of peptidic over metallic origin. Nevertheless, since X-PCI-CT phase maps are directly related to δ, i.e. the refractive index decrement in the Xray complex index of refraction formula n = 1-δ + iβ, with δ proportional to the local electron density ρe, and ρe = Z/V = Z *ρ *Na/A (Z atomic number, V volume of reference, ρ macroscopic mass density, NA Avogadro's number and A atomic weight), an equivalent volumetric amount of a high-Z material, such as Fe, will contribute almost twice to X-PCI hyper-intensity compared to a low-Z material such as P (ZFe/ZP=1.73). For this reason, the atomic properties of Fe make it a key player in the generation of the X-PCI-CT-observed ICHD signals.

Impact of sample shrinkage on data quantification:
Moderate systematic dehydration-related sample shrinkage, either due to sample-preparation (fixation, paraffination) or during X-PCI-CT acquisitions, may account for the slight discrepancy between literature and measured HD-particle sizes. Since all extracted brains were subject to the same sample-preparation workflow and X-ray irradiation imaging protocols, the comparative analysis of particle size between different animal groups still holds. Actually, 3D tissue virtual analyses (such as the X-PCI-CT-based one applied here), which avoid the lossy traditional histological approach (often leading to partial section analysis and interpolation) hold the promise of more complete and trustworthy neuropathological evaluations compared to more traditional approaches.

More on the results of the cellular-level quantification:
Also the measurements of HD-particle sizes generally matched literature values for cellular sizes in the mouse brain: diameters of nucleoli, glia and normal neuron somas are reportedly around