Microglial Expression of the Wnt Signaling Modulator DKK2 Differs between Human Alzheimer’s Disease Brains and Mouse Neurodegeneration Models

Abstract Wnt signaling is crucial for synapse and cognitive function. Indeed, deficient Wnt signaling is causally related to increased expression of DKK1, an endogenous negative Wnt regulator, and synapse loss, both of which likely contribute to cognitive decline in Alzheimer’s disease (AD). Increasingly, AD research efforts have probed the neuroinflammatory role of microglia, the resident immune cells of the CNS, which have furthermore been shown to be modulated by Wnt signaling. The DKK1 homolog DKK2 has been previously identified as an activated response and/or disease-associated microglia (DAM/ARM) gene in a mouse model of AD. Here, we performed a detailed analysis of DKK2 in mouse models of neurodegeneration, and in human AD brain. In APP/PS1 and APPNL-G-F AD mouse model brains as well as in SOD1G93A ALS mouse model spinal cords, but not in control littermates, we demonstrated significant microgliosis and microglial Dkk2 mRNA upregulation in a disease-stage-dependent manner. In the AD models, these DAM/ARM Dkk2+ microglia preferentially accumulated close to βAmyloid plaques. Furthermore, recombinant DKK2 treatment of rat hippocampal primary neurons blocked WNT7a-induced dendritic spine and synapse formation, indicative of an anti-synaptic effect similar to that of DKK1. In stark contrast, no such microglial DKK2 upregulation was detected in the postmortem human frontal cortex from individuals diagnosed with AD or pathologic aging. In summary, the difference in microglial expression of the DAM/ARM gene DKK2 between mouse models and human AD brain highlights the increasingly recognized limitations of using mouse models to recapitulate facets of human neurodegenerative disease.


Introduction
Microglia, the resident immune cells of the CNS, contribute both beneficially and detrimentally to Alzheimer's disease (AD) in a context-dependent manner, thus rendering their response to AD heterogeneous in nature. So too is their phenotype at the transcriptomic, proteomic, epigenomic, metabolomic, and morphologic level leading to the identification of spatiotemporally distinct microglial subpopulations (for review, see Masuda et al., 2020;Paolicelli et al., 2022). Disease-associated (DAM) or activated response microglia (ARM; henceforth: DAM/ ARM) represent a subpopulation associated with the neurodegenerative brain (Keren-Shaul et al., 2017;Sala Frigerio et al., 2019). Transitioning from homeostatic to DAM/ARMstate requires TREM2 (triggering receptor expressed on myeloid cells-2; Keren-Shaul et al., 2017). A bona fide receptor for b Amyloid, TREM2 ligation activates microglia and orchestrates a gene regulatory response that increases inflammatory signaling, phagocytosis, and proliferation, a response thought to restrict development of AD (for review, see Gratuze et al., 2018).
TREM2 regulates microglial proliferation and survival by activating, among others, the canonical Wnt/b -catenin pathway (Zheng et al., 2017; for review, see Aghaizu et al., 2020). Indeed, several genes upregulated by TREM2 in response to AD pathology are related to proliferation and Wnt signaling (Meilandt et al., 2020). The canonical Wnt signaling modulatory gene Dkk2 (Mao and Niehrs, 2003) was upregulated downstream of Trem2 in DAM/ARM cells in APP/PS1, PS2APP, 5xFAD, and APP NL-G-F AD mouse models in separate studies, making it a putative DAM/ARM marker gene (Friedman et al., 2018;Sala Frigerio et al., 2019;Meilandt et al., 2020). Database searches further indicate that while control CNS DKK2/ Dkk2 expression levels are generally low across the various cell types, they are, respectively, either similar or greater in nonmicroglial CNS cell populations compared with microglia in human and mouse single-cell RNA sequencing (RNA-Seq) studies (Zhang et al., 2014(Zhang et al., , 2016Friedman et al., 2018). The secreted protein DKK2 belongs to the Dickkopf family of Wnt modulators (Niehrs, 2006). Its homolog DKK1 antagonizes Wnt signaling through Frizzled Wnt receptors by sequestering the co-receptor LRP5/6 (Bafico et al., 2001;Mao et al., 2001). The reduced Wnt signaling tone evident in AD is at least partially because of Ab fibril-induced upregulation of DKK1/Dkk1 in human AD and AD mouse models (Caricasole et al., 2004;Rosi et al., 2010;Killick et al., 2014;Sellers et al., 2018;Jackson et al., 2019). This was synaptotoxic in in vitro and in vivo models (Purro et al., 2012;Galli et al., 2014;Marzo et al., 2016;Elliott et al., 2018;Sellers et al., 2018), and potentially also in human AD (Jackson et al., 2019).
Much less is known about the role of microglial DKK2 in the CNS, not to mention in AD. In cell lines, DKK2 can both antagonize and agonize Wnt-LRP6 signaling depending respectively on the presence or absence of the second co-receptor Kremen2 (Mao and Niehrs, 2003). During neural crest specification, DKK2 agonizes Wnt signaling (Devotta et al., 2018). Conversely, in cancer studies DKK2 generally inhibits Wnt signaling (Kuphal et al., 2006;Sato et al., 2007;Maehata et al., 2008;Hirata et al., 2009;Zhu et al., 2012;Mu et al., 2017). Furthermore, cancer cell-secreted DKK2 suppresses immune cell activation via an unconventional Wnt-unrelated pathway (Xiao et al., 2018). In the aforementioned single-cell and bulk cell gene expression studies on neurodegeneration mouse models, Dkk2 was upregulated in microglia, but no information on the spatial relationship between Dkk2 1 microglia and neurodegenerative pathology or the biological role of this upregulation was provided (Friedman et al., 2018;Sala Frigerio et al., 2019;Meilandt et al., 2020). To address this gap in our knowledge, we performed a histologic assessment of microglial Dkk2/DKK2 upregulation in several mouse models and in human AD and furthermore investigated the effect of recombinant DKK2 on cultured primary neurons.
Here, we report significant microgliosis and microglial Dkk2 mRNA upregulation in a disease-stage-dependent manner in APP/PS1, and APP NL-G-F AD mouse model brains. Clustering of Dkk2 1 microglia around amyloid plaques was often more pronounced than that of Dkk2microglia. In cultured rat neurons, recombinant DKK2 blocked Wnt-dependent synapse formation. Crucially however, microglial DKK2 upregulation was not detected in postmortem human brain from individuals diagnosed with AD or pathologic aging. This nonuniversality of what was a putative DAM/ARM marker gene highlights the increasingly recognized limitations of using animal models to recapitulate facets of human neurodegenerative disease.
In every case, mice were housed according to the appropriate institution's ethical requirements, and in compliance to the country's laws for animal research. Typically, mice were housed in standard individually ventilated cages with less than or equal to three mice per cage at 21 6 1°C with relative humidity 55 6 10% and maintained on a 12/12 h light/dark cycle with access to food (standard pellets), water, and nesting material provided ad libitum via an overhead rack. At the onset of pathology, affected animals were provided with food pellets soaked in water at ground level to ensure sufficient nourishment and hydration. Cages were checked daily to ensure animal welfare. Body weight was assessed regularly to ensure no weight loss. For animals housed at WuXi AppTec, studies were reviewed and approved by Institutional Animal Care and Use Committee (IACUC) of WuXi AppTec (Suzhou) Co, Ltd. For animals housed at VIB/kU Leuven, studies were approved by the kU Leuven Ethical Committee and in accordance with European Directive 2010/63/EU. For animals housed at UCL, studies were conducted following the guidelines of the UCL Institute of Neurology Genetic Manipulation and Ethic Committees and in accordance with the European Community Council Directive of November 24, 1986 (86/ 609/EEC). Animal experiments were undertaken under license from the United Kingdom Home Office in accordance with the Animals (Scientific Procedures) Act 1986 (Amended Regulations 2012) and were approved by the Ethical Review Panel of the Institute of Neurology.
For tissue collection, animals were injected with terminal anesthesia (pentobarbital sodium, Euthatal) and were transcardially perfused with PBS by trained personnel.

Rats
Animal experiments were undertaken under license from the United Kingdom Home Office in accordance with the Animals (Scientific Procedures) Act 1986 (Amended Regulations 2012) and in compliance with the ethical standards at University College London (UCL). Timed matings were set up for Sprague Dawley rats for subsequent harvesting of embryos at embryonic day (E)18. Pregnant rat dams were killed using Isoflurane and cervical dislocation.

Human postmortem tissue
Anonymized human samples from control, pathologic aging, and AD subjects were obtained from the Queen's Square Brain Bank for Neurologic Disorders (QSBB) and NeuroResource, UCL Institute of Neurology, University College London. All samples were obtained with informed consent in accordance with the Human Tissue Act 2004 and under the UCL Institute of Neurology HTA material transfer agreement UCLMTA1/17 approved by the NHS Research Ethics Committee. Postmortem frontal cortex biopsy tissue was harvested, snap-frozen, and stored at À80°C until further tissue processing. All experiments were performed in accordance with relevant guidelines and regulations. Sample information including demographic data, disease classifications and postmortem intervals is shown in Extended Data Table 6-1.

Tissue processing
Mouse brain tissue, freshly harvested on transcardial perfusion with PBS, was postfixed by immersion in 4% paraformaldehyde (PFA) in PBS overnight at 4°C followed by overnight immersion in and equilibration to 20% sucrose in PBS at 4°C. Mouse brains were split into three segments by applying 2 equidistant coronal slice cuts along the rostro-caudal axis, resulting in an olfactory bulb containing rostral-most segment, a hippocampus containing middle segment and the caudal-most cerebellar segment. After embedding in OCT (CellPath) and freezing in 2-methylbutane (Sigma) prechilled in liquid Nitrogen, the middle segment was coronally cryosectioned at 15mm thickness on a Leica CM1860UV cryostat (Leica) and sections containing clearly defined hippocampus were transferred onto Superfrost Plus Gold microscopy slides (ThermoScientific).
Mouse spinal cord tissue was processed identically but split only into two segments by applying a transverse slice cut rostral to the lumbar enlargement, resulting in a rostral cervical/thoracic segment and a caudal lumbar segment. The cryo-embedded lumbar segment was transversally cryosectioned at 15-mm thickness and sections containing clearly defined L5 lumbar spinal cord were transferred onto Superfrost Plus Gold microscopy slides (ThermoScientific).
Human frontal cortex brain tissue was cryosectioned at 15-mm thickness, sections were transferred onto Superfrost Plus Gold microscopy slides (ThermoScientific) and dried for 10 min at 40°C.
Human and mouse sections were stored at À80°C until staining.
mRNA fluorescence in situ hybridization (FISH) mRNA fluorescence in situ hybridization for mouse Dkk2 and human DKK2, TREM2, and P2RY12 mRNA was performed on mouse brain/spinal cord and human frontal cortex cryosections respectively, by using the Multiplex Fluorescent V2 Assay kit (ACD Bio).
Briefly, mouse cryosections were thawed and dried at 40°C for 4 min before postfixation with 4% PFA in PBS at room temperature (RT) for 10 min. OCT residue was washed off by applying 1Â PBS for 5 min at RT. Sections were treated with RNAScope H 2 O 2 for 4 min at RT and subsequently washed 2Â 3 min with UltraPure Distilled Water (Invitrogen) at RT. Microscopy slides containing cryosections were submerged for 4 min in boiling 1Â RNAScope target retrieval solution followed by immediate submersion in UltraPure Distilled Water. Cryosections were dehydrated in 100% ethanol at RT for 2 min and allowed to air dry at RT for 5 min. Cryosections were subsequently treated with RNAScope Protease IV at RT for 15 min and washed 2Â 3 min at RT with 1Â PBS. RNAScope probes were allowed to hybridize to cryosections for 2 h at 40°C (Mm-Dkk2-C1, 404841; Mm-Ppib-C1 (positive control probe), 313911; E. coli-Dapb-C1 (negative control probe), 310043). Probes were detected with TSA-Cy3 (PerkinElmer, FP1170) using the RNAScope branched DNA amplification principle as per the manufacturer's instructions. Subsequently, cryosections were further immunohistochemically processed (see below).
Neuronal transfection with the DNA construct pHR hsyn:EGFP [Keaveney et al., 2018; kind gift from Xue Han (Addgene plasmid #114215; http://n2t.net/addgene: 114215; RRID:Addgene_114215)] was performed at 7 d in vitro (DIV) using the Neuromag magnetofection method (OzBiosciences). Briefly, for every 40,000 cells plated per well of an eight-well chamber slide dish, 0.5-mg DNA was mixed and complexed with 1-ml Neuromag transfection reagent in 100 ml of OptiMem (all reagents at room temperature). Following 20 min of incubation at room temperature, the transfection mix was added dropwise to neuronal cultures and the culture dish was placed on a magnetic plate (OzBiosciences) preequilibrated to 37°C inside an incubator for the magnetofection step. After 20 min of magnetofection in the incubator, cell culture dish was removed from the magnetic plate and normal cell culture resumed.
Fixation was performed following 24 h of recombinant protein treatment using 4% PFA/4% sucrose (Sigma) in 1Â PBS at RT for 15 min. Neurons were subsequently washed 3Â with 1Â PBS.

Microscopy
Stained tissue was imaged using a Zeiss LSM 880 confocal laser scanning microscope fitted with 40Â (NA = 1.3) and 63Â (NA = 1.4) objectives and photomultiplier tubes to detect fluorescence emission. For image acquisition, xyz confocal stacks were captured at a resolution of 1024 Â 1024 pixels and at a step size of 1 mm. Microscope settings were established during first acquisition and subsequently not further modified. Four distinct fields of view (FOVs) were imaged from two representative sections per sample.
For image acquisition of transfected primary rat hippocampal neurons following recombinant protein treatment, whole neurons were acquired using the 40Â objective and secondary dendrites were acquired with higher magnification using the 63Â objective with an additional 3.5Â zoom at a resolution of 1024 Â 1024 pixels and at a step size of 0.5 mm. A total of 15 neurons and accompanying secondary dendrites spread across three biological repeats were imaged per condition.

Image processing and analysis
All images acquired from mouse tissue were processed and analyzed in Fiji/ImageJ (Schindelin et al., 2012). xyz confocal stacks were collapsed into maximum z projections. Microgliosis was assessed by measuring both the number of microglia (DAPI 1 nuclei embedded within typical microglial Iba1 immunoreactivity) and the total 2D surface area of Iba1 immunoreactivity within the acquired field of view. For area quantification, Iba1 immunoreactivity was processed by applying the "Remove Outliers" function to remove nonspecific noise (bright, radius = 2, threshold = 50), followed by thresholding at 35/255 to define the signal range, and two further rounds of the "Remove Outliers" function, to fill-in nuclear and other gaps in Iba1 staining (dark, radius = 1), and to further remove nonspecific noise (bright, radius = 3). The created Iba1 surface area was measured and used as a mask within which the Dkk2 mRNA FISH signal surface area, thresholded to 30/255, was quantified. Normalized Dkk2 area per microglial cell was determined by dividing the total measured Dkk2 area by the number of detected Iba1 1 /DAPI 1 microglia within a given field of view.
Human frontal cortex image acquisitions were first subjected to "Linear unmixing" with automatic fluorophore detection within the Zeiss Zen Black software (Zeiss) to remove overlapping signals between the five fluorophore channels. Unmixed and maximum z projected images were subsequently processed and analyzed using the HALO FISH-IF v2.0.4 module (Indica Labs). The DKK2 mRNA FISH signal surface area associated with TREM2/ P2RY12 double positive microglia cells was quantified. To achieve this, cell nuclei and their xy coordinates were recorded based on DAPI signal. Probe detection was optimized based on signal size, intensity of positive probe pixels and contrast threshold parameter settings (see Extended Data Table 6-2). The maximum distance threshold for probe signal assignment to nuclei was 25 mm. We classified cells positive for P2RY12 and TREM2 as microglia (DAPI 1 /P2RY12/TREM2 1 ), determined their number, and measured the surface area of DKK2 mRNA FISH signal associated with such DAPI 1 /P2RY12/TREM2 1 cells. Normalized DKK2 area per microglial cell was determined by dividing the total measured DKK2 area by the number of detected DAPI 1 /P2RY12/TREM2 1 microglia within a given field of view.
Microglia-b Amyolid plaque distance analysis: we determined the 2D Euclidian distance of microglia to the proximal most b Amyolid plaque dense core in maximum projected images according to the following Fiji/ImageJ methodology: an intensity threshold was applied to the image channel containing b Amyloid immunostaining to identify the plaque dense core, which was usually more intensely labeled compared with the plaque periphery; because of the heterogeneous nature of b Amyloid plaques, threshold values were determined for each acquired image. In early-stage APP/PS1, and APP NL-G-F AD mouse or littermate control tissue devoid of b Amyolid plaques, plaque dense core "placeholders" were randomly placed on confocal images by digitally drawing appropriately dimensioned white ellipses on the color channel assigned to b Amyloid immunostaining using Fiji/ImageJ, followed by intensity threshold application as above. The binary dense core image generated in the previous step was subjected to the "Exact Signed Euclidian Distance Transform (3D)" (EDT) plug-in to create a 2D map where distance to the closest dense core was encoded in gray values from À1024 (furthest possible distance) to 0 (at dense core edge). xy position landmarks of DAPI 1 microglia nuclear centers were placed on a binary image, which in turn was redirected to the EDT image in the "Set Measurements" window, selecting "Mean gray value" as measurement output. Note that xy positions of human microglia exported from HALO FISH-IF v2.0.4 module were imported into Fiji/ ImageJ using the macro "ImportXYcoordinates.ijm." Gray values at microglial xy positions were obtained using the "Analyze Particles" function and converted into distance units by multiplying the gray value by the image xy pixel dimension (0.13495 mm) to yield microglia-b Amyloid plaque distances.
Dendritic spine and synapse analysis on hSyn:EGFPexpressing primary rat hippocampal neurons following recombinant protein treatment was performed using IMARIS software. Briefly, the "Filament" tool was used to semi-automatically specify the secondary dendrite within an image file, followed by the detection of dendritic spines by manual identification. Postsynaptic Homer immunoreactivity usually manifested as puncta in dendrites, especially within dendritic spines. To quantify the number of Homer puncta exclusively within the transfected secondary dendrite of interest, the GFP signal was used to create an exclusion mask using the "Surface" tool to isolate the Homer signal within the transfected dendrite. Homer puncta were subsequently identified using the "Spot" detection tool set to a detection diameter of 0.45 mm; background Homer signal was excluded by thresholding using the "Quality" filter. Presynaptic vGlut puncta in the entire image were similarly identified using the "Spot" detection tool at 0.45-mm diameter. Synapses were assumed using the "Colocalize spots" function within the "Spot" detection tool when there was a maximum distance of 1 mm between Homer and vGlut puncta.

Experimental design and statistical analysis
All means are stated 6 SD. For the histologic study aspects, N = number of subjects (humans or animals) and n = number of fields of view. For qualitative and quantitative histologic assessments, we typically examined at least four subjects per group, imaging at least four different fields of view from two cryosections per subject, which met previously conducted sample size calculations according to Rosner (2015) with data inputs from Friedman et al. (2018). For the cytological study aspects, N = number of biological repeats, n = number of technical repeats (cells analyzed). We used GraphPad Prism software (GraphPad Software Inc.) for statistical analyses. D'Agostino and Pearson test was used to assess the normality of datasets. For the comparison of one independent variable between more than two groups, we used one-way ANOVA with Tukey's multiple comparison test. For statistical tests involving two independent variables we used two-way ANOVA with Sidák multiple comparisons test; where data points were missing, Mixed-effects analysis with Sidák multiple comparisons test was utilized. Significance was accepted at p 0.05 (see Table 1; alphabetical superscripts in results section and figure legends refer to Table 1).

Data, software, and code availability
The datasets generated during and/or analyzed during the current study are available from the corresponding authors on request. The Fiji/ImageJ macro "ImportXYcoordinates. ijm" is available on the Github repository available via https:// github.com/DominicAghaizu/ImageJMacros/blob/main/ ImportXYcoordinates.ijm.

Results
Microgliosis and microglial Dkk2 upregulation in APP NL-G-F mice We first investigated the microglial Dkk2 expression pattern in the APP NL-G-F knock-in AD mouse model,  (Saito et al., 2014). To this end, we performed mRNA FISH on coronal brain cryosections to detect Dkk2 mRNA in situ and acquired images from the motor cortex and the stratum pyramidale, with adjacent stratum oriens and stratum radiatum, of the hippocampal CA1 region, which are brain regions burdened by b Amyloid plaques, neurofibrillary tangles and neuronal degeneration in AD patients and animal models. This was paired with immunohistochemical labeling using antibodies against Iba1 and b Amyloid to assess microglial Dkk2 expression, as suggested previously (Friedman et al., 2018;Sala Frigerio et al., 2019;Meilandt et al., 2020), and to evaluate the spatial relationship between microglia and b Amyloid plaque lesions. As expected, the brains of wild-type control littermate mice at 7 or 24 months were devoid of b Amyloid plaques and exhibited normally tiled Iba1 1 microglia (Fig. 1A,B). In stark contrast, we detected b Amyloid plaques in the cortex and CA1 of age-matched transgenic APP NL-G-F mice at 7 and 24 months (Fig. 1A,B). This was accompanied by robust microgliosis as assessed by both normalized microglia cell count (DAPI 1 /Iba1 1 cells) and area of Iba1 signal in maximum z-projected image stacks ( Fig. 1C; note that the microglia spatial distribution will be addressed below). In the cortex, the number of microglia was significantly higher in APP NL-G-F mice relative to age-matched littermate controls at seven months [6.5 6 0.8 vs 3.0 6 0.2 microglia per field of view (FOV; equal to 1.8 Â 10 À2 mm 2 )] and at 24 months (15.9 6 3.8 vs 2.8 6 0.8 to); significant differences were found between time points and genotypes (Fig. 1A,D,G; two-way ANOVA, p = 0.0101 and p = 0.0002, respectively a ). Iba1 area was also significantly elevated in transgenic mice compared with littermate controls, both at seven months (1505.8 6 135.0 vs 821.9 6 239.3 mm 2 ) and 24 months [1982.4 6 471.6 vs 781.7 6 121.1 mm 2 ; Fig. 1A,E; two-way ANOVA, p = 0.1821 (time points) and p = 0.0018 (genotypes) b ]. To assess microglial Dkk2 expression levels, we quantified Dkk2 mRNA FISH signal that was colocalized with Iba1 immunoreactivity (Fig. 1C). The normalized area of Dkk2 signal per DAPI 1 / Iba1 1 microglial cell reached significantly higher levels in APP NL-G-F mice relative to littermate controls, both at seven months (0.3 6 0.2 vs 0.1 6 0.1 mm 2 ) and at 24 months [1.2 6 0.4 vs 0.1 6 0.1 mm 2 ; Fig. 1A,F,G; two-way ANOVA, p = 0.0245 (time points) and p = 0.0013 (genotypes) c ].
Similar patterns of microgliosis and Dkk2 upregulation were observed in the hippocampal CA1 region. Microglia count numbers were markedly elevated in APP NL-G-F mice compared with littermate controls both at seven months Finally, we also assessed whether there were genotype-related changes in the relative contribution of Dkk2 1 microglia versus the total microglia population. In the cortex, the percentage of Dkk2 1 microglia was significantly elevated in APP NL-G-F mice compared with littermate controls both at seven months (48.1 6 22.3 vs 18.8 6 5.1%) and at 24 months [84.1 6 5.1 vs 15.6 6 7.7%; Extended Data Fig. 1-1A; two-way ANOVA, p = 0.0240 (time points) and p = 0.0004 (genotypes) g ]. In the CA1 hippocampus, the percentage of Dkk2 1 microglia was similarly elevated in APP NL-G-F mice compared with littermate controls at seven months [34.9 6 16.9 vs 7.8 6 9.4%) and at 24 months (58.4 6 21.6 vs 18.4 6 17.7%; Extended Data Fig. 1-1B; two-way ANOVA, p = 0.0784 (time points) and p = 0.0093 (genotypes) h ].
Taken together, our data demonstrate robust microgliosis in conjunction with Dkk2 upregulation in APP NL-G-F mice compared with littermate controls, adding a spatial dimension to a previously published single-cell RNA sequencing (RNA-Seq) study that identified Dkk2 expression in DAM/ARM microglia of the same mouse model .
Microgliosis and microglial Dkk2 upregulation in APP/ PS1 mice Following investigation of APP NL-G-F mice, we assessed microgliosis and Dkk2 upregulation in a second AD mouse model, the APP/PS1 mouse, that expresses chimeric mutant mouse/human App and mutant human presenilin 1, both associated with early onset familial AD in humans (Jankowsky et al., 2004).
b Amyloid plaque load progressively increased in APP/ PS1 mice starting from eight months, whereas agematched wild-type control littermates lacked b Amyloid plaques altogether. This was especially evident in the cortex ( Fig. 2A). While plaques were detectable in the hippocampus of APP/PS1 mice (data not shown), CA1 stratum pyramidale proximal regions, the standardized hippocampal brain region that was imaged in our study, rarely exhibited plaque depositions (Fig. 2B).
As noted above, hippocampal CA1 stratum pyramidale proximal regions in APP/PS1 mice were mostly devoid of b Amyloid plaques. Here, we were unable to detect any changes in the number of DAPI 1 /Iba1 1 microglia (Fig. 2B, G,J), Iba1 area (Fig. 2B,H), and Dkk2 mRNA signal per microglial cell compared with age-matched littermate controls (Fig. 2B,I,J; two-way ANOVA, all n.s.).
Finally, unlike in APP NL-G-F mice, we could not detect any significant time point-related changes in the relative contribution of Dkk2 1 microglia versus the total microglia population in APP/PS1 mice compared with littermate controls both in the cortex [Extended Data Thus, we were able to largely replicate our findings regarding microgliosis and microglial Dkk2 upregulation in two widely used AD mouse models (APP NL-G-F and APP/ PS1 mice), again adding spatial information to a previously published meta-analysis of single-cell RNA-Seq datasets (Friedman et al., 2018). However, the lack of b Amyloid plaques and microglial phenotype in hippocampal CA1 stratum pyramidale proximal regions of the APP/PS1 mouse evokes the notion that the microglial phenotype investigated here could be linked to plaque proximity.

Dkk2 1 microglia exhibit increased propensity for clustering around bAmyloid plaques
To investigate whether Dkk2 expression status was correlated with b Amyloid plaque proximity, we performed nearest neighbor analysis to quantify the spatial relationship between microglia and the nearest b Amyloid plaque dense core identified following b Amyloid IHC in APP NL-G-F and APP/PS1 mice (schematic shown in Fig.  3A). Frequency distributions of recorded distances were summarized in histograms. In AD mouse models, we distinguished between Dkk2 1 and Dkk2microglia, whereas no such distinction was made in wild-type mice as Dkk2 expression levels were negligible at all time points (Figs. 1A,B,F,J, 2A,B,E,I). Furthermore, where no plaques were evident (e.g., in wild-type or predisease stage mice or in some hippocampal CA1 stratum pyramidale proximal regions) distances to plaque dense core "placeholders" randomly placed on confocal images were measured instead.
As would be expected, wild-type littermate controls of APP NL-G-F mice used in our study exhibited microglia at varying/random distances to the nearest randomly assigned dense core placeholder in the motor cortex and CA1 hippocampus at 7 and 24 months (Figs. 1A,B, 3B-E). This finding is in keeping with the homogeneous tiling behavior usually exhibited by microglia in the healthy CNS (Nimmerjahn et al., 2005). In stark contrast, a large proportion of Dkk2 1 and Dkk2microglia were found within 20 mm of the nearest plaque dense core in the cortex of seven-month-old APP NL-G-F mice, while Dkk2 1 microglia were predominantly located within 40 mm of plaque dense cores in the CA1 hippocampus (Fig. 3B,D). By 24 months, the clustering of microglia around b Amyloid plaque dense cores, especially that of Dkk2 1 microglia, became even more pronounced both in the cortex and in the CA1 hippocampus (Fig. 3C,D). Skewness and kurtosis analyses of histogram distribution curves for each individual animal revealed that APP NL-G-F microglia were statistically significantly more tightly clustered around plaque dense cores with increasing age than microglia of age-matched control mice (Extended Data Fig. 3-1A-H). Crucially, however, in 24-month-old APP NL-G-F mice, Dkk2 1 microglia were statistically significantly more tightly associated with plaques than Dkk2microglia both in the cortex and in the CA1 hippocampus (Extended Data Fig. 3-1C,D,G,H; oneway ANOVA n ).
We observed similar plaque-microglia distance relationships in APP/PS1 mice and respective wild-type littermate controls. Microglia in the wild-type littermate control mouse cortex and CA1 hippocampus were evenly distributed relative to the nearest randomly placed plaque dense core placeholder (Fig. 3F-K). In the cortex of three-  month-old (predisease stage and plaque free) APP/PS1 mice, both Dkk2 1 and Dkk2microglia exhibited similar wild-type-like distance distributions (Fig. 3F), whereas microglia increasingly clustered within 20 mm of plaque dense cores at subsequent (disease stage) time points, with Dkk2 1 microglia exhibiting slightly more pronounced clustering versus Dkk2microglia at 12 months (Fig. 3G, H); we note that the latter difference was not statistically significant according to skewness and kurtosis analysis, while clustering of Dkk2 1 microglia around plaques dense cores in APP/PS1 mice was statistically significantly increased versus that of microglia of age-matched control mice from eight months onwards (Extended Data Fig. 3-1I-N; one-way ANOVA o ). As discussed above, because of the small amounts of b Amyloid plaques in hippocampal CA1 stratum pyramidale proximal regions of the APP/PS1 mouse, microglia distributions were comparatively variable, especially at three months (Fig. 3I), although substantial clustering of Dkk2 1 microglia was registered in those instances were b Amyloid plaques were observed in CA1 stratum pyramidale proximal regions at 8 and 12 months (Fig. 3J,K). Accordingly, skewness and kurtosis analyses were inconclusive for CA1 microglia (Extended Data Fig. 3-1O-T; one-way ANOVA p ).
While it is widely known that microglia accumulate around CNS lesions such as b Amyloid plaques, our data further suggest that clustering around plaques is frequently accompanied by the expression of Dkk2, especially in the APP NL-G-F AD mouse model. Conversely, in the healthy brain, microglia were evenly tiled and lacked Dkk2 expression.

Microgliosis and microglial Dkk2 upregulation in SOD1 G93A ALS mice
Having demonstrated microgliosis and clustering of Dkk2 1 microglia around b Amyloid plaques in two different widely used AD mouse models, we next investigated whether our findings could be recapitulated in another neurodegeneration mouse model, the SOD1 G93A amyotrophic lateral sclerosis (ALS) mouse (Gurney et al., 1994). According to the meta-analysis of single-cell RNA-Seq datasets by Friedman et al., 2018; microglial Dkk2 upregulation should also be evident in this mouse model. It expresses the mutant human SOD1 G93A gene that causes motor neuron degeneration in the spinal cord and other parts of the CNS, which underlies ALS (Gurney et al., 1994). We performed mRNA FISH to detect microglial Dkk2 mRNA in situ paired with immunohistochemical labeling using an antibody against Iba1 on transverse cryosections from the lumbar (L)5 region of the spinal cord and acquired images from the ventral horn, an area that displays robust motor neuron degeneration in this mouse model (Gurney et al., 1994).
We next sought to investigate whether microgliosis and microglial Dkk2 upregulation in the SOD1 G93A ALS mouse model were spatially correlated with local CNS lesions, analogous to that observed in the APP NL-G-F and APP/PS1 AD mouse models. In absence of AD-typical b Amyloid plaques in ALS, we combined Dkk2 mRNA FISH and microglial immunohistochemical labeling with the immunolabelling of GFAP to visualize astrocytes and immunolabelling of misfolded SOD1 to visualize aggregates of misfolded mutant SOD1 G93A . In 120-d-old SOD1 G93A ALS mice, we failed to detect clustering of microglia, regardless of their Dkk2 expression status, specifically around GFAP (Fig. 4F). However, we observed some degree of continued hippocampus. G, Quantification of microglia numbers per maximum projected FOV. H, Iba1 IHC surface area per maximum projected FOV. I, Normalized Dkk2 mRNA FISH signal area per DAPI 1 /Iba 1 microglial cell. J, Comparative % changes of Dkk2 expression and microglia numbers during time course. Individual data points represent the average of four FOVs analyzed for each animal (C-E, G-I) or total averages from all animals per group (F, J). N = 6 animals (3Â females, 3Â males) per time point and condition, n = 4 different fields of view/animal and brain region. Scale bars: 25 mm (A-C) and 5 mm (magnified ROIs). Data show mean +/-SD. Two-way ANOVA with multiple comparisons test. *p , 0.05, **p , 0.01, ***p , 0.001, ****p , 0.0001 ( i-k ). See also Extended Data Figure 2-1. However, many microglia did not exhibit local accumulation around misfolded SOD1 lesions (Fig. 4G,H, cyan ROIs). In absence of a clear clustering pattern, these observations were not quantified.
Taken together, the microgliosis and microglial Dkk2 upregulation detected in the brains of AD mouse models could also be replicated in an unrelated neurodegeneration mouse model, namely in the spinal cord of the SOD1 G93A ALS mice. While some degree of clustering around misfolded SOD1 aggregates occurred, this was not as robust as the clustering around b Amyloid plaques in the APP NL-G-F and APP/PS1 AD mouse models. Nonetheless, our findings support the published notion that Dkk2 upregulation may be part of a general response in CNS microglia as they transition from surveillance to activation (DAM/ARM microglia), at least in mouse models of neurodegeneration (Friedman et al., 2018;Sala Frigerio et al., 2019;Meilandt et al., 2020). This supports the possibility that Dkk2 represents a DAM/ARM marker gene, at least in mice.

DKK2 recombinant protein disrupts WNT7a-induced synapse features in cultured neurons
We next sought to investigate what effect Dkk2 protein secreted by microglia might have on its surroundings under the assumption that increased microglial Dkk2 expression at the mRNA level results in increased microglial Dkk2 protein secretion. We focused our study on synapses in mature primary neuron cultures because of the well-known anti-synaptic effect that the Dkk2 homolog Dkk1 has on them, which it brings about by decreasing canonical and increasing noncanonical Wnt signaling (Purro et al., 2012;Galli et al., 2014;Marzo et al., 2016;Elliott et al., 2018;Sellers et al., 2018). However, we note that, in principle, Dkk2 can have context-dependent agonistic and antagonistic effects (Mao and Niehrs, 2003).
To this end, we treated mature rat hippocampal neuron cultures sparsely expressing hSyn:EGFP at 21 d in vitro with recombinant proteins for 24 h (WNT7a, 200 ng/ml; DKK1, 100 ng/ml; DKK2, 100 ng/ml; DKK2 1 WNT7a, 100 and 200 ng/ml; BSA control, 100 ng/ml). Chosen recombinant protein concentrations were in line with published works and/or TCF/LEF dose dependence assays performed in house (data not shown). This was followed by immunocytochemical labeling using antibodies against the presynaptic and postsynaptic markers vGlut and Homer. A typical sparsely labeled (hsyn:EGFP 1 ) neuron with highlighted primary dendrite (boxed ROI) that was used for analysis is depicted in Figure 5A. WNT7a treatment significantly increased the number of dendritic spines as well as the number of postsynaptic homer puncta compared with BSA treatment (Fig. 5B,C,G,H; one-way ANOVA, dendritic spines: p = 0.0023 u ; homer puncta: p = 0.0309 v ). Conversely, these metrics were unaffected by DKK1 and DKK2 treatment (Fig. 5D,E,G, H; one-way ANOVA; all n.s.) and crucially also by combined DKK2 1 WNT7a treatment (Fig. 5F-H; one-way ANOVA; n.s.). The absolute number of synapses (defined as Homer/vGlut apposition events with up to 1 mm distance) was similarly increased by WNT7a but not by DKK1, DKK2 or a combination of DKK2 and WNT7a compared with BSA, although this did not reach statistical significance ( Fig. 5I; one-way ANOVA; n.s.).
Nonetheless, these combined data suggest that DKK2 treatment is antagonistic rather than agonistic and completely abolishes the pro-synaptogenic effect of WNT7a treatment, at least in our in vitro assay. Furthermore, it appears that the antagonistic effect of DKK2 as well as that of the established Wnt signaling antagonist DKK1 rely on an inherent Wnt signaling tone that was low/absent in our cultures, as neither reduced synaptic metrics to levels below those found with BSA treatment when applied independently.

DKK2 is not upregulated in human microglia
We have thus far demonstrated significant microgliosis and microglial Dkk2 upregulation in AD and ALS mouse models of neurodegeneration, as well as clustering of Dkk2 1 microglia around b Amyloid plaques. In combination with previously published studies, which have demonstrated microglial Dkk2 upregulation by singlecell RNA-Seq (Friedman et al., 2018;Meilandt et al., 2020), this led us to postulate that Dkk2 may represent a bona fide DAM/ARM marker gene at least in neurodegeneration mouse models. We next sought to investigate whether our findings were recapitulated in human subjects diagnosed with AD.
We subsequently assessed the clustering behavior of microglia around b Amyloid plaques. In healthy control individuals, the total microglia population displayed a varying/random spatial distribution around the nearest randomly placed dense core placeholder, which furthermore did not appear to be modified by DKK2 expression status (Fig. 6A,F). In individuals classified as "pathologic aging," we identified emerging populations of both DKK2 1 and DKK2microglia that frequently accumulated around b Amyloid plaque dense cores up to a distance of 50 mm, although clustering in the proximal most regions was more robust for DKK2cells (Fig. 6B, G). This clustering was further consolidated, especially among DKK2 1 microglia, whose predominant distribution now also included proximal most regions (Fig. 6C,H).
Taken together, our data on human frontal cortex postmortem tissue indicate that neither the increase in microglial numbers nor microglial DKK2 upregulation, both of which were evident in mouse models, occur in human brains under conditions classified as "pathologic aging" and "AD." However, microglia did exhibit clustering behavior around b Amyloid plaques although this was not linked to DKK2 expression.

Discussion
Past and present research have linked dysregulated Wnt signaling to AD (Palomer et al., 2022; for previous review, see Purro et al., 2014;Palomer et al., 2019;Aghaizu et al., 2020;Inestrosa et al., 2021). However, recent research has also more intimately linked microglia and neuroinflammation to AD, as initially exemplified by variants of genes predominantly expressed in microglia like TREM2 and CD33 exhibiting disease modifying properties (Bradshaw et al., 2013;Guerreiro et al., 2013). New evidence even suggests that the microglial AD response is itself regulated by Wnt signaling, as the signaling pathway downstream of TREM2, essential for regulating microglial survival and proliferation, cross-talks with the Wnt pathway (Zheng et al., 2017;Meilandt et al., 2020).
Here, we sought to explore the role of DKK2/Dkk2, which encodes a Wnt signaling modulator, that was upregulated in a subpopulation of microglia (DAM/ARM) in various single and bulk cell RNA-Seq studies on neurodegeneration mouse models (Friedman et al., 2018;Sala Frigerio et al., 2019;Meilandt et al., 2020). Our histologic data obtained largely by mRNA FISH combined with immunocytochemistry replicated the findings cited above. Crucially however, we added valuable spatial information on the location of Dkk2 1 microglia with respect to neurodegenerative lesions such as b Amyloid plaques in AD mouse models, where Dkk2 1 microglia exhibited a potential to cluster near b Amyloid plaques that was greater or at least equal to that of Dkk2microglia, at least in advanced-APP NL-G-F mice. The exact role of Dkk2 protein expression is yet to be fully understood, but assuming its reported role as a secreted, soluble protein (for review, see Niehrs, 2006), we speculated that Dkk2's mechanism of action could be autocrine or paracrine in nature. In support of the former, oncological evidence suggests that peripheral immune natural killer and CD8 1 T cells, which are derived from the same myeloid lineage as CNS microglia, can detect soluble Dkk2. However, in this context, Dkk2 was utilized as an immune evasion tool secreted by tumors to suppress cytotoxic immune cell activation and tumor destruction via an atypical, Wnt signaling independent pathway (Xiao et al., 2018). Nonetheless, it is a possibility that microglial-derived Dkk2 can also act on microglia in an autocrine fashion at least in mice, although we can only speculate what the cellular response to such a stimulus would be.
Conversely, we provide evidence in support of a paracrine mechanism at least in cultured rat primary neurons as we demonstrate that recombinant human DKK2 protein blocks the synaptogenic effect of Wnt in vitro. However, we note that the administered DKK2 protein continued from all animals per group (E). N = 3 animals per time point and condition, n = 4 fields of view per animal. Scale bars: 25 mm (A, F-H) and 5 mm (magnified ROIs). Data show mean +/-SD. Two-way ANOVA with multiple comparisons test. *p , 0.05, **p , 0.01, ***p , 0.001, ****p , 0.0001 ( q-s ). See also Extended Data Figure 4-1. Figure 5. Recombinant DKK2 protein neutralises the synaptogenic effect of WNT7a in mature hippocampal primary neurons. A, Typical rat hippocampal neuron at DIV22 expressing hSyn:EGFP immunolabelled with Homer and vGlut. Boxed ROI indicates a primary dendritic branch, on which analysis in this section was focused. B-F, Representative primary dendrites of DIV22 hippocampal neurons treated for 24 h with 100 ng/ml BSA control (B), 200 ng/ml WNT7a (C), 100 ng/ml DKK1 (D), 100 ng/ml DKK2 (E), and 100 ng/ml/200 ng/ml DKK2 1 WNT7a (F). Immunolabelling for the presynaptic and postsynaptic markers vGlut and homer was performed, and merged views are shown in top panels. Remaining panels show homer (middle panel) and vGlut (lower panel) with outlined primary dendrite boundaries based on hSyn:EGFP labeling. G, Normalized number of dendritic spines per 100-mm primary dendrite. H, Normalized number of homer puncta per 100-mm primary dendrite. I, Normalized number of synapses (defined as homer/vGlut apposition events with a maximum distance of 1 mm) per 100-mm primary dendrite. N = 3 biological repeats, n = at least 15 analyzed neurons per condition. Scale bars: 50 mm (A) and 5 mm (B-F). Data show mean +/-SD. One-way ANOVA with Tukey's post hoc test. *p , 0.05, **p , 0.01 ( u, v ). concentration may not match physiological, microglia derived Dkk2/DKK2 protein levels in situ. Knowing that DKK2 can generally engage in Wnt antagonizing and agonizing activities depending respectively on the presence or absence of the co-receptor Kremen2 (Mao and Niehrs, 2003), it appears that, at least in our in vitro system, DKK2 protein acts as an antagonist. DKK2 may thus behave similarly to DKK1, a negative regulator of canonical Wnt/b -catenin and noncanonical Wnt/PCP signaling with known synapse destabilizing properties Figure 6. DKK2 is not upregulated at the mRNA level in postmortem brains from AD patients. A-C, Representative confocal images depicting microglial DKK2 expression in the human frontal cortex. DKK2 as well as microglial TREM2 and P2RY12 mRNA FISH signal in conjunction with b Amyloid IHC labeling in postmortem human frontal cortex samples from healthy control individuals (A), individuals diagnosed with pathologic aging (B), and individuals diagnosed with AD (C). Boxed ROIs highlight microglia expressing DKK2 (DAPI 1 /DKK2 1 /TREM2 1 /P2RY12 1 ); yellow boxed ROIs were enlarged for improved visualization. D, Quantification of microglia (DAPI 1 /TREM2 1 /P2RY12 1 ) numbers per maximum projected FOV (FOV = 1.8 Â 10 À2 mm 2 ). E, Normalized DKK2 mRNA FISH signal area per DAPI 1 /TREM2 1 /P2RY12 1 microglial cell. F-H, Distribution of DAPI 1 /TREM2 1 /P2RY12 1 microglia [Dkk2 1 , Dkk2 -, or total microglia (MG) populations] distances to nearest b Amyloid plaque dense core in postmortem human frontal cortex samples. Individual plots show relative frequency distributions in individuals classified as healthy control (F), pathologic aging (G), and AD (H). Healthy control individuals: N = 5 individuals, n = 8 fields of view); AD (Braak & Braak stage 5-6): N = 6 individuals, n = 8 fields of view; pathologic aging (Braak & Braak stage 3-4): N = 2 individuals, n = 8 fields of view. Data points represent the average of four FOVs analyzed for each individual subject (mean 6 SD); individual subject mean values were further averaged for each group of interest and summarized as mean 6 SD (blue horizontal bars, red error bars). One-way ANOVA with Tukey's post hoc test ( w, x ). No statistical differences identified. Scale bars: 25 mm (A-C) and 5 mm (A-C enlarged ROIs). See also Extended Data Figure 6-1, Extended Data Tables 6-1 and 6-2. (Purro et al., 2012;Galli et al., 2014;Killick et al., 2014;Marzo et al., 2016;Elliott et al., 2018;Sellers et al., 2018; also, for review, see Aghaizu et al., 2020), likely also in the human AD brain (Caricasole et al., 2004). Synapse density reductions in plaque proximal regions (Koffie et al., 2009) would be consistent with the fact that oligomeric b Amyloid induces Dkk1 expression (Purro et al., 2012;Killick et al., 2014;Jackson et al., 2019). Dkk2 1 microglia accumulating around b Amyloid plaques may locally increase Dkk2 protein levels, adding to the anti-synaptic milieu established by Dkk1 near plaques. Given that microglia already engage in complement-mediated synaptic pruning by phagocytosis in AD mouse models (Hong et al., 2016;Shi et al., 2017), the relative contributions of individual synaptotoxic components around plaques will have to be addressed in future studies.
In assessing the chronological order between microgliosis/microglial plaque clustering and microglial Dkk2 upregulation, we observed significant microgliosis increases before Dkk2 upregulation in APP/PS1, APP NL-G-F , and SOD1 G93A mice with respect to absolute quantification metrics (Figs. 1, 2, 4). However, when comparing relative rate changes, the rate of Dkk2 signal increase at early disease stages in the APP/PS1 AD mouse model surpassed the rate of microgliosis increase (Fig. 2F,J). It should be noted that Dkk2 induction was initiated from near-zero basal expression levels (Fig. 2E), whereas both basal microglia numbers and Iba1 immunoreactivity levels were decidedly greater than zero (Fig. 2C,D,G,H). The potential for more pronounced changes was thus markedly greater for Dkk2 induction at least in APP/PS1 mice. Conversely, in SOD1 G93A ALS mice, the rate of microgliosis increase surpassed that of Dkk2 signal increase at early disease stages (Fig. 4E). Presumably, basal microglial cell densities lower than those observed in the mouse brain (Figs. 4B,C vs 2C,D,G,H), which is in keeping with published literature (Tan et al., 2020), likely contributed at least partially to this outcome.
What should be addressed in future studies is whether the ability to induce Dkk2 expression is innate in all microglia or whether context, such as proximity to neurodegenerative lesions, is to be ascribed a more prominent role. CNS microglia are not a homogeneous population of cells, with gene expression signatures differing depending on factors such as brain region, sex, age, and context including disease (for review, see Masuda et al., 2020). In three-month-old APP NL-G-F mice, Dkk2 1 ARM cells represented 6% of to the total microglial pool ; this number increased to 33% and 52% at 6 and 12 months, respectively. It will be interesting to discern whether ARM-competence is restricted to the initial population of ARM cells, which then serve as a proliferating seed population, or whether cells from the total microglia pool are continuously recruited into the Dkk2 1 ARM population as disease progresses. The potential to produce Dkk2 1 ARM may further by influenced by other factors, which should be addressed in future studies, as different neurodegeneration disease models and CNS regions analyzed in our study exhibited varying contributions of Dkk2 1 microglia relative to the total microglial pool.
Finally, our study has revealed discrepancies between human AD and transgenic AD mouse models. DKK2 mRNA expression levels were not elevated in postmortem frontal cortex samples from individuals diagnosed with AD versus healthy individuals. While other human brain and CNS regions like the motor cortex, hippocampus, and spinal cord might exhibit DKK2 upregulation (although unlikely given the absence of DKK2 upregulation in recently published human RNA-Seq databases; see below), the above finding is in stark contrast to our findings in neurodegeneration mouse models. In contrast to the situation in human patients, proximity to b Amyloid plaques appeared to be a strong predictor of microglial Dkk2 expression in mice, both in the hippocampus and motor cortex. We note that those microglia that exhibited DKK2 expression at the mRNA level in human tissue also displayed higher levels of the microglial marker P2RY12, but not TREM2. While the relevance of this finding is yet to be determined, published research has shown microglial expression of P2RY12, typically considered a homeostatic microglial marker gene, in proximity to diffuse plaques in postmortem tissue from AD individuals (Walker et al., 2020). A caveat worth mentioning in relation to the lack of DKK2 upregulation is the fact that human microglia at AD end stage (Braak & Braak stage 5-6) were chronically exposed to disease for much longer periods than their mouse counterparts and chronic adaptations in microglia gene expression signatures as well as microglial numbers may have masked potential earlier changes (we note that our pathologic aging samples at Braak & Braak stage 3-4 also lacked DKK2 upregulation). Nevertheless, it is now known that gene expression signatures between mouse and human DAM/ARM populations, although overlapping to some extent, exhibit distinct differences (for review, see Wang, 2021). In fact, numerous single-cell RNA-Seq analyses have identified gene expression signatures that differed between mouse and human DAM/ARM populations (Grubman et al., 2019;Mathys et al., 2019;Nguyen et al., 2020;Olah et al., 2020;Smith et al., 2022). For technical reasons and in contrast to mouse studies, human single-cell RNA-Seq studies are frequently, although not exclusively (Olah et al., 2020), restricted to nuclear transcripts, which may have contributed to the apparent transcriptomic differences between mouse and human microglia (note that extra-nuclear mRNA is abundant because of nuclear export before translation). However, even in a recent single nucleus RNA-Seq comparative study involving human AD postmortem tissue and the 5xFAD AD mouse model, differences between human and mouse microglial gene expression signatures persisted (Zhou et al., 2020). The inability to detect extra-nuclear mRNA in human brain samples can be circumvented with the use of optimized tissue harvesting protocols (Olah et al., 2020), or in situ detection methods such as low throughput mRNA FISH (present study; Jolly et al., 2019) or higher throughput digital spatial profiling (Prokop et al., 2019). Nonetheless, our mRNA FISH based study strengthens the notion that human and mouse microglia, despite exhibiting some overlaps, are different even beyond just the expression status of DKK2/Dkk2, at least in the brain. Future studies should also examine any such interspecies differences in the spinal cord.
Our study therefore highlights the increasingly recognized difficulties and limitations of using mouse models to recapitulate facets of human biology and disease (Elder et al., 2010;Jucker, 2010;Cavanaugh et al., 2014;Justice and Dhillon, 2016;Perlman, 2016;Dawson et al., 2018). Regardless of whether this may be ascribed in our study to differing biological responses in humans versus mice or masking chronic adaptations in much longer human disease, these limitations likely play a key role in the absence of truly disease altering therapies to date despite decades of AD research and .100 clinical trials. Future AD research should thus substantially increase scrutiny in cases where animal models are to be used to ensure faithful modeling of human biology. Human based AD models including human induced pluripotent stem cell-derived cell cultures and brain organoids are potent additions to our tool-kit despite still lacking the capacity to fully recapitulate human in vivo biology in an in vitro setting, and indeed in an in vivo setting .