Abi3 regulates microglial ramification and dynamic tissue surveillance in vivo

A rare coding variant of Abelson-interactor gene family member 3 (Abi3) is associated with increased risk of late-onset Alzheimer’s Disease (AD). Although Abi3 is recognised as a core microglial gene, its role in microglia is largely unknown. Here we demonstrate that Abi3 is crucial for normal microglial morphology, distribution, and homeostatic tissue surveillance activity in vivo.

3 the WAVE2 complex 14 and exogenous ABI3 expression reduced WAVE2 translocation to the cell periphery 15,16 by preventing c-ABL/WAVE2 phosphorylation 15 . Ectopic expression of ABI3 in NIH3T3 fibroblasts, which endogenously express extremely low levels of ABI3 14 , has been shown to impair cell spreading 14 . In a cancer cell line ABI3 forced overexpression caused cell cycle arrest in G0/G1 phase 17 , while in cultured rat hippocampal neurons both overexpression and knock-down of Abi3 caused changes in dendritic spine morphogenesis and synapse formation 18 . Moreover, ABI3 was found to be enriched in microglial clusters around Aβ-plaques in AD brains 19 and recent whole-exome microarray analyses identified a rare ABI3 coding variant, p.Ser209Phe, as a risk factor for AD development 20 , prompting a need to understand its role in microglia, a cell type which has been implicated in AD 21 . Taken together, these observations suggest a role for ABI3 in microglia in AD, which has not yet been explored. Given the requirement for a physiologically-relevant microenvironment for microglial phenotype, this study focused on the study of cells of the macrophage lineage and microglia in vivo, both with endogenous Abi3 expression levels.
Macrophage motility, as well as their phagocytic activity, depend on the ability of these cells to correctly form lamellipodia and filopodia structures. Interestingly, ABI3 overexpression in NIH3T3 fibroblasts altered the formation of lamellipodial and actin-mediated cell spreading 14 . To examine Abi3 role in a more physiologically-relevant context, we generated conditionally-immortalised macrophage precursor (MØP) cell lines by transduction of CD117 + bone marrow cells from age and gender matched Abi3-WT and -KO mice with a retrovirus encoding oestrogen-dependent version of Hoxb8 22 . Abi3 deficiency was confirmed at genomic level in both Abi3-KO mice and MØP cell lines and by qPCR of macrophages derived from the MØPs (Supplementary Fig. 1a-b). Macrophages differentiated from MØPs in M-CSF resemble bone-marrow derived macrophages 22 and exhibited increased Abi3 expression in Abi3-WT cells, but expression was not evident in the Abi3-KO cells, as 4 expected ( Supplementary Fig. 1c-d). The MØP-derived macrophages enabled examination of cell spreading in an appropriate cell type as well as in the context of physiologically-relevant endogenous Abi3 expression levels. The ability of Abi3-WT and -KO macrophages to spread on fibronectin-coated coverslips was tested using an actin-mediated spreading assay. Cells were examined at four time points over a 4-hour period and analysed ( Supplementary Fig. 2) to obtain area and solidity values for the spreading of each cell. Abi3-deficient macrophages showed a marked increase in cell size compared to Abi3-WT control cells, with a noticeable reduction in the number of ramifications (Fig. 1a). Indeed, Abi3-WT cells had significantly smaller surface and were richer in processes, with a resulting decrease in cellular solidity, at all time-points (Fig. 1b-i and Supplementary Table1), even up to a week after re-plating ( Supplementary Fig. 3). Similar results were obtained with two additional pairs of independently generated cell lines, confirming that the observed phenotype was linked to genotype and not an artifact of an individual cell line.  (15,30,120  Given the marked phenotypic differences observed between Abi3-WT and -KO macrophages and the high expression of Abi3 in microglia, we hypothesized that Abi3 could similarly control microglial morphology and associated functionality. As microglia are intrinsically connected to astrocytes and neurons, any perturbation of this equilibrium could have additional repercussions, especially in a pathological brain microenvironment, as observed in AD; for this reason, we examined the microglia of the animals in situ. Microglial morphology and distribution were assessed in the Prefrontal Cortex (PFC) and the Hippocampus of young (8-week-old) Abi3-WT and -KO mice due to the contribution of these two brain regions to memory formation and retention 23,24 and their involvement in AD pathogenesis 25,26 . In accordance with the in vitro macrophage data, Abi3-KO microglia stained with an Iba1 antibody lacked the complex ramifications of Abi3-WT cells, and consistently showed shorter and thicker processes in both brain regions (Fig. 2a). Abi3-KO mice also showed a significant increase in microglia number and concomitant reduced Nearest Neighbour Distance (NND) in the PFC (Fig. 2b-c) as well as in the Hippocampus ( Fig. 2e-f) when compared to the Abi3-WT mice. However, the Iba1 + area of brain tissue was decreased compared to the Abi3-WT controls in both PFC (Fig. 2d) and Hippocampus (Fig   2g) despite the increased cell density, suggesting an impairment in microglia ability to efficiently monitor brain parenchyma. 6 The gross phenotype observed in Abi3-KO mice has been reported as indicative of activated microglia. CD68 is a lysosomal marker upregulated in phagocytic microglia and regarded as a marker of microglia activation 27 . For this reason, we investigated CD68 levels in the absence of Abi3. Interestingly, histological analysis of Abi3-WT and -KO mice brains highlighted a significant increase in CD68 levels within microglia in the absence of Abi3 in both the hippocampus (Fig. 2h, j) and PFC (Fig. 2i).
Activated microglia have been shown to release multiple factorsincluding proinflammatory cytokines, Nitric Oxide, Transforming Growth Factor α and Vascular Endothelial Growth Factor Bwhich in turn regulate astrocytic phenotype 28 . Thus, while Abi3 is mainly expressed by microglia 11 , the loss of Abi3 could indirectly impact astrocytes.
We therefore assessed Glial Fibrillary Acidic Protein (GFAP) positive cells within the Hippocampus of young (8-week-old) Abi3-KO mice. We observed an increase in the number of GFAP + astrocytes within the hippocampus of Abi3-KO mice ( Fig. 2k-l), while no significant difference was evident in the levels of GFAP immunoreactivity (Fig. 2m), suggesting a potential alteration of astrocytic morphology following loss of Abi3. To examine in detail the extent of the morphological abnormalities presented by Abi3-KO microglia, we traced the 3-dimensional structure of single microglia (Fig.3a) and analysed the resultant data using a linear mixed model (Supplementary Table 2  These histological findings pointed to a likely alteration of microglia surveillance activity, due to the marked decreased in ramification complexity in the absence of Abi3. Microglial surface protrusions are extremely dynamic and are constantly scanning the surrounding parenchyma, and this can be visualised in vivo by 2-photon microscopy through cranial windows in awake mice 29 . Abi3-WT and -KO mice were crossed with a Cx3cr1-EGFP reporter line 30 , to generate mice heterozygous for the EGFP reporter. Two-photon analysis on unchallenged awake mice revealed that Abi3-KO microglia exhibited reduced surveillance of brain parenchyma, with only 51.88% ± 7.49% and 53.32% ± 3.77% (for males and females respectively; mean ± SD) of the total area covered by motile microglial processes during the 45-minute acquisition period, compared to 75.41% ± 1.80% and 73.07%  To date, the physiological function of Abi3 in macrophages, and specifically microglia, was largely unknown. Previous in vitro studies implicated it in the regulation of actin dynamics [14][15][16]18,31 . This is the first report to address the role of Abi3 in microglia in vivo, demonstrating its control of both morphology and function of these cells.
The dystrophic phenotype observed in the Abi3-KO mice, together with their increased CD68 levels, is of particular interest in the pathological context of AD, where microglia have been implicated in Aβ clearance through phagocytosis 21 , and an impairment in microglia migration towards amyloid plaques has been shown to significantly worsen amyloid burden 21 .
Moreover, alterations in microglial phenotype could affect their synaptic pruning activity, possibly leading to the increased synapse loss observed in AD 32 . These observations highlight 11 the need for further investigation of the role of microglial motility and tissue surveillance in the risk of AD.
In conclusion, the data presented here support the hypothesis that Abi3, a core microglial gene, contributes to AD development through its fundamental role in the regulation of microglial morphology and movement, which alters homeostatic surveillance of brain tissue by microglia. This could in turn cause an impairment in microglial phagocytic activitywith potentially detrimental effects on debris removal and Aβ clearanceas well as synaptic pruning. Therefore, our studies identify Abi3 as a key molecule of interest for further investigation in AD.

Corresponding authors
Correspondence to Elena Simonazzi or Philip R. Taylor.

Competing interests
The authors declare no competing interests. Common-Rv CCCAGACACTCGTTGTCCTT. The final PCR products were 410 bp for WT mice and 500 bp for Cx3cr1-GFP mutant mice.

Generation of conditionally-immortalized Macrophage Precursor (MØP) cell lines
Hoxb8 conditionally-immortalised macrophage precursor (MØP) cell lines were derived from CD117 + enriched bone-marrow cells taken from individual 8week-old mice, euthanised with CO2 asphyxiation, using established methods 22   Coverslips were imaged using a Zeiss Apotome Axio Observer microscope (Zeiss) or an EVOS™ FL Auto 2 Imaging System (Thermo Scientific) microscope with a 20x objective.
Slides were independently blinded and 15-30 random fields were analysed for each experimental group (10 for each coverslip) using Fiji software (Fiji is just ImageJ, version 1.52p 33 ). After a first step of brightness and contrast optimisation, manual separation of cells in close contact was performed using a 2 μm-thick black line. After carefully adjusting the threshold and generating a binary file, the "Fill holes" tool automatically filled any gap within the particles, which were then analysed using the "Analyse Particles" function. Due to the expected ramified morphology of macrophages, the two chosen parameters for the following statistical analysis were cell area and cell solidity (described by the

Immunofluorescent staining
Coronal sections from two regions of interest, the Prefrontal Cortex (between 3.08 mm to 2.58 mm from the bregma) and the Hippocampus (-1.34 mm to -2.18 mm from the bregma), were selected for each mouse using Images from the Mouse Brain atlas 34

Image acquisition and analysis
Brain sections were imaged using a Zeiss Cell Observer Spinning Disk confocal microscope (Zeiss) with a 20x objective. Eight z-stacks were acquired for the PFC and 14 for the Hippocampus region. The central 30 optical sections of each z-stacks were transformed into a frontal maximum-intensity orthogonal projection and used for the following analyses with Fiji Software (Fiji is just ImageJ, version 1.52p 33 ). After utilising the "Background subtraction" tool, Iba1 or GFAP positive cells were manually counted using the "Cell Counter" plugin. Care was taken not to include cells with their cell body partially outside the edges of the image. Each image was then converted to the 8-bit format and Iba1 or GFAP positive pixels were selected by applying a threshold to generate a binary image. The "Measure" tool allowed the evaluation of area percentage covered by Iba1 + or GFAP + pixels.
In the case of the microglial staining, the 8-bit image was further processed to obtain Nearest Neighbour Distance (NND) values. Two sequential filters from the MorphoLibJ plugin library were applied to the images: a Gray Scale Attribute opening filter (Operation = "Opening", Attribute = "Area", Minimum Value = 25 pixels, Connectivity = 8) was applied to isolate the cells from the background, followed by an opening Morphological filter (Operation = "Opening", Element = "Octagon", Radius = 1 pixel) in order to separate the cell body from the ramifications. After generating binary images using the "Entropy Threshold" plugin, missing cell bodies or wrongly included ramifications were manually corrected to reduce bias. Centroids were calculated for each cell body selecting a 30µm minimum size filter in the "Analyse Particles" tool. Finally, the "NND" plugin was used to calculate the relative values for all the identified cell bodies. Experiments were performed in blinded conditions, in order to avoid any operator bias.
For CD68 and Iba1 co-localisation analysis, the area percentage covered by Iba1 + cells was used to generate a ROI, that was then applied to the CD68 staining (processed in the same manner as the Iba1 channel to obtain a binary image) before using the "Measure" tool, in order to make sure only colocalizing pixels were comprised in the analysis.

Morphometric analysis of single cells
Iba1 + cells in the CA1 region of the Hippocampus were images using a Zeiss Cell Observer Spinning Disk confocal microscope (Zeiss) with a 40x oil immersion objective. Between 4 20 and 6 z-stacks of random fields were acquired for each mouse. Single cells were carefully cropped out of each image taking care to select only cells that were entirely comprised within the z-stack and whose cell body was not overlapping with adjacent cells. A total of 5 cells per animal was then analysed using the Filament Tracer module of Imaris software (Bitplane, Oxford Instruments, version 9.3.1). The filament starting point was placed at the soma and the seed point threshold was adjusted to include all the ramifications. Manual editing of the algorithm-generated filament was performed by an operator blinded to the experimental condition, in order to remove or correct ill-traced ramifications and to add any branch missed by the algorithm. The numerical output of the following parameters was extracted for further statistical analyses: branch length, branch mean diameter, branch volume, total number of branching points, full branch level, cell area, cell volume and Sholl intersections.

Two-photon microscopy
A week before the imaging session, 8-9 weeks old Abi3-WT and -KO mice crossed to B6.129P2(Cg)-Cx3cr1 tm1Litt /J mice underwent a craniotomy as described by Goldey et al. 29 .
Briefly, a 3 mm-large circular portion of skull centred over the primary visual cortex (3.00 mm lateral to lambda and 1.64 mm anterior to the lambdoid suture 29 ) was removed and replaced with two 3 mm cover glasses and one 5mm cover glass on the top, and a metal head plate fixed to the skull by dental cement. A week after the surgery mice were image, d using a resonant scanning two-photon microscope (Thorlabs, B-Scope) with a 16x 0.8 NA objective with 3-mm working distance (Nikon). Mice were awake during the imaging session and positioned on a cylindrical treadmill. Laser power was set to 30mW (at the sample) for all recordings. Each mouse underwent a single imaging session, during which a total of 30 zstacks (comprised of 25 optical sections, divided by a 2µm step size for a total z volume of 50 μm) of the same random area were acquired every 90 seconds over a period of 45 minutes 21 using a 7x zoom in order to ensure visualisation of the thin processes. Each focal plane of the z-stack was imaged 10 consecutive times before moving to the next one, which then allowed image registration to be performed with a custom written software in MATLAB ® (MathWorks). Each z-stack was then merged in a maximum intensity orthogonal projection and then collated with all the others, in chronological order.
The resulting file was then used to perform a quantification of the area covered by Abi3-WT and -KO microglia using Fiji software. A single orthogonal projection was obtained flattening the time-stack. A first threshold was applied to select all the ramifications, while a second one was used to only comprise the brightest objects, which mostly encompassed the cell bodies. The percentage of "dynamic area"i.e. the area covered only by dynamic structureswas measured after subtracting the second binary image from the first one.

Statistics
Statistical analyses were mostly performed using GraphPad Prism (GraphPad Software,  Table. Two-way ANOVA tests for genotype and sex factors were used for basic microglia, CD68 and astrocyte characterisation, as well as for the data generated from the analysis of the two-photon videos. Data generated from Imaris were analysed using linear mixed-model analysis in R (version 4.0.0). To account for possible confounders, in addition to the genotype the regression models included sex and mice IDs, as well as radius or branch level wherever appropriate; genotype, sex and radius/branch level were included as fixed effects, while the animal's ID was included as a random effect to account for the animalspecific variability.