Microglial transcriptional profiles of a transgenic rat model closely model Alzheimer's disease

Abstract

Single-cell RNA-sequencing has identified that Alzheimer's disease (AD) pathology in humans is associated with activation of disease-associated microglia (DAM). Microglial signatures of human AD have not been consistently identified in AD mouse models. Since the inflammatory response of rats is more like humans, we profiled microglial transcriptomes in aging TgF344-AD rats, which overexpress two human AD risk genes. Classic DAM gene activation (ApoE, Trem2, Gpnmb), and upregulation (MHC class-II) and downregulation (Ifngr1 and Fkbp5) of human AD microglial genes were identified in aging TgF344-AD rats. Thus, the TgF344-AD rat better recapitulates the microglial gene signature observed in human AD.

Introduction

Single-cell transcriptomic studies of Alzheimer's disease (AD) mouse models have revealed that protein aggregation and neuronal damage are associated with activation of disease-associated microglia (DAM). ¹ Microglia, the resident macrophages of the brain and spinal cord, are dynamic immune cells poised to adapt to challenges in the local environment. ² In AD mice, DAM display downregulation of “homeostatic” microglial genes (P2ry12, Cx3cr1, Cd33, Tmem119). ² Initially, DAM are neuroprotective, expressing genes involved in lipid metabolism (Apoe, Lpl, Tyrobp, Trem2). ¹ These neuroprotective DAM engulf debris and misfolded proteins, including amyloid-β (Aβ).^1,3 However, as Aβ and/or tau pathology progresses, DAM become neurotoxic, with upregulation of genes involved in the innate immune defense system (C3, Tspo, Cxcl16).^4,5 These have recently been classified as terminally inflammatory microglia (TIM). ⁶ With AD progression, the fragile balance between neuroprotective DAM and TIM is consequently disrupted. Thus, the potential to modulate dynamic DAM responses provides an unprecedented opportunity to alter the progression of AD.

A subset of dysregulated DAM genes are shared between human AD and Aβ mouse models of AD, including upregulation of TREM2, CD68, and APOE.^7–10 However, other DAM microglia signatures from mouse models of AD have not been consistently identified in human AD, suggesting species-independent cellular responses.^7–9,11,12 The heterogenous expression of DAM signature genes in AD may reflect differing profiles of two subgroups of AD-associated microglia that were recently identified: (1) Aβ-associated microglia expressing DAM genes, and (2) tauopathy-associated microglia expressing the gene for the glutamate receptor GRID2. ⁹ The DAM microglial signature in human AD has been suggested to be driven by IRF8. ⁷ Thus, an animal model that more fully recapitulates the microglial gene expression changes that occur with AD is required. Using single-cell RNA-sequencing (scRNA-seq) of hippocampal microglia, we demonstrate here that the TgF344-AD rat, which overexpresses human mutant APPSW and presenilin 1 (PS1ΔE9), ¹³ better recapitulates the Aβ-driven microglial response in human AD than mouse models. Top differentially expressed transcripts (Tmem119 and Gpnmb) were further evaluated using RNA in situ hybridization, confirming upregulation of Gpnmb and a trend towards downregulation of Tmem119.

Methods

Animals

All animal procedures were approved by the University of California-Davis Institutional Animal Care and Use Committees (protocol #22337). Female TgF344-AD rats and wild-type littermates (n = 4 per group) were sampled at either 3 or 10 months of age. The 10-month time point was selected because, by this age, TgF344-AD rats display increased Aβ plaque load, increased Iba1/Cd68 expression and cognitive deficits, relative to age- and sex-matched congenic controls. ¹³ Tau pathology, detected by antibodies against specific phospho-tau epitopes (AT8, CP13, pS199/202, pS396) and conformation-selective tau antibodies (MCI and PHF1) stained neurons in hippocampi and cortices are not apparent until 16 months of age. ¹³ Breeding and genotyping were performed as previously described.^14,15 At 3 or 10 months of age, female rats were anesthetized with 4% isoflurane in a 2:1 mix of medical grade air and oxygen, flowing at a rate of 1.5 L/min. Transcardial perfusion was performed with ice-cold PBS. Rats were euthanized via exsanguination under anesthesia.

Hippocampal microglia isolation and scRNA-seq

Hippocampi were dissociated into single cells as described previously for dorsal root ganglia. ¹⁶ Cells were suspended in PBS with a 1% FCS buffer (or BSA, aka “FACS buffer”), and staining occurred to the resuspended pellet in 100 µl of residual FACS buffer for 30 min at room temperature. For cell staining, cells were incubated with an anti-mouse CD45 antibody (Clone: OX1, ThermoFisher, Catalog# 11-0461-82, RRID AB_2572455, 0.25 μg per million cells in 100 μL) conjugated to FITC and an anti-rat CD11b/c antibody (Clone: OX-42, BioLegend, Catalog t# 201809, RRID AB_313995, 0.25 µg per million cells in 100 µL) conjugated to APC. Stained cells were washed in FACS buffer 1× to remove unbound Maps at 1200 rpm in a tabletop centrifuge. Cells were analyzed and sorted immediately thereafter using the Beckman Coulter “MoFlo Astrios EQ”. Cells were gated at CD45^low and CD11b⁺, while excluding Gr-1 positive cells and doublets. Two animals were pooled per group prior to library preparation to maximize the amount of microglia obtained. Barcoded 3′ single-cell libraries were prepared using the Chromium Single Cell 3′ Library and Gel Bead kit v2 (10X Genomics). Libraries were sequenced on an Illumina HiSeq4000, with paired-end 100 bp reads to a targeted median depth of 250,000 reads per cell. ¹⁷

Bioinformatic analysis

Cellranger v.2.0.1 and bcl2fastq v.2.17.1.14 commands mkfastq and count were used to generate fastq files per sample, align to Rnor6.0, filter, and perform barcode and unique molecule identifier (UMI) counting. Cells were filtered out if they contained <200 genes, demonstrated mitochondrial gene expression >15% and/or had UMIs <200 or >1.8 × 10⁴. Normalization, clustering, and calculation of UMAP coordinates was conducted using Seurat, version 4.1.0 in R. ¹⁸ Initial cell type identification was conducted by mapping to the Allen Mouse Brain Atlas (http://mouse.brain-map.org/) to distinguish astrocytes, dentate gyrus granule cells and microglia-perivascular macrophages. Specific microglia markers (P2ry12 and Hexb) and perivascular macrophage markers (Cd163 and Mrc1) where then used to determine the main microglia cluster (Supplemental Figure 1) and clusters were confirmed using scMRMA. ¹⁹ All downstream analyses were performed in microglia only. Differential expression (DE) analyses between genotypes and age, adjusting for sample, was conducted on the filtered and normalized data using limma, version 3.50.3. ²⁰ GO enrichment analysis of DE results was conducted with the Bioconductor package topGO in R, version 2.46.0, as previously described. ¹⁶ Significance was set at a Benjamini-Hochberg false discovery rate adjusted p-value (p_FDR) of <0.05.

RNA in-situ hybridization

After perfusion, the right hemisphere of the rat brains were cut into 2-mm coronal sections. These were fixed in 4% wt/vol paraformaldehyde (PFA; sigma Chemical) in 0.1 M phosphate buffer for 24 h at 4°C, and then transferred to 30% wt/vol sucrose (ThermoFisher Scientific) in PBS for at least three days before being embedded in Optimal Cutting Temperature (OCT) compound (ThermoFisher Scientific) and flash frozen in a methanol and dry ice bath. These blocks were crytosectioned into 10 μm sections of fixed frozen rat brain tissue mounted on Superfrost Plus slides (Fisher Scientific). Samples for single-molecule fluorescence in situ hybridization (ISH) were prepared from AD affected regions: (1) hippocampus: three regions including CA1, CA3 and the dentate gyrus and (2) the entorhinal cortex from rats in each experimental group at 3 and 10 months of age (n = 2–3 female rats per group) as previously described.^16,21 We performed duplex chromogenic ISH manually on the slides using the RNAscope 2.5 HD duplex reagent kit (Cat#322430, Advanced Cell Diagnostics, Inc. (ACD)). Target probes were Rn-Tmem119-norvegicus transmembrane protein 119 (Tmem119, Cat# 478921) mRNA and Rn-Gpnmb-C2-Rattus norvegicus glycoprotein nmb (Gpnmb, Cat# 553791) mRNA (ACD). Controls used for validation of signal included positive control probe (Rn) PPIB-C1/ POLR2A-C2 (ACD Cat# 320821) and negative control probe dapB (ACD Cat# 310043) (Supplemental Figure 2).

Section peroxide pretreatment and target retrieval was performed according to the manufacturer's recommendations (ACD TN 320534). Briefly, slides were rinsed in 1X PBS to remove OCT, baked for 30 min at 60°C and post fixed with prechilled 4% PFA for 15 min at 4°C. Sections were dehydrated with 50%, 70%, 100% ethanol for 5 min each at RT and then allowed to air dry. Sections were treated with hydrogen peroxide for 10 min at RT, rinsed briefly in water, and then immersed in 1X target retrieval buffer held at 98–102°C for 5 min. Slides were rinsed with water and 100% ethanol and allowed to air dry. The sections were treated with protease plus reagent and incubated in a humidified chamber for 30 min at 40°C. Probe hybridization, amplification and signal detection were carried out according to the manufacturer's recommendations (ACD 322500-USM). After probe hybridization, the slides were rinsed twice in wash buffer and held overnight at RT in 5X SSC buffer. Sections were counterstained in 50% hematoxylin, rinsed quickly in water and then immersed in 0.02% ammonia water. Slides were dried in the hybridization oven for 15 min at 60°C, cooled, and then mounted with VectaMount media (Vector Labs H-5000).

Slides were imaged with an Olympus VS120 Slide Scanning System and analyzed using ImageJ. ²² RNA expression was quantified by performing a color deconvolution of the brightfield RGB images to separate the stains into different channels, auto-thresholding each channel using the Intermodes method to separate foreground and background, and measuring the % area stained by the red (Gpnmb) and blue (Tmem119) labels within each region of interest. Three regions of interest were quantified in the hippocampus (CA1, CA3 and the dentate gyrus), in addition to the entorhinal cortex. Analysis was performed blinded to the corresponding transcript for each colored label. Quantification was compiled from all four regions, found to be normally distributed using a Shapiro-Wilk test, and analyzed using a 2-way ANOVA, with genotype, age and genotype×age interaction as variables. Multiple comparisons were performed using Tukey's multiple comparison tests, and an adjusted p value of <0.05 was considered significant. All data analyses were performed using GraphPad Prism software (Boston, MA USA).

Microglial validation

In order to confirm that the RNA signal obtained from our ISH was from microglia and no other cell types, subsequent immunofluorescence staining for Iba1 for microglia, and NeuN for neurons, was performed. Two sections from the hippocampus of each rat (Bregma −3.3 to −4.2; validated using The Rat Brain in Stereotaxic Coordinates) were used. Slides underwent antigen retrieval in sodium citrate buffer for 30 min at 95°C, and immunostained with primary antibodies diluted in serum blocking buffer overnight at 4°C: anti-Iba1 (1:1000, Wako Chemicals, 019-19741) and NeuN (1:500, Millipore, MAB377). Secondary antibodies were incubated for an hour at room temperature: AlexaFluor 568 (1:500, Life Technologies, A11036) and AlexaFluor 488 (1:500, Invitrogen, A11017). Slides were then mounted in ProLong Gold with DAPI (ThermoFisher Scientific). Negative controls were incubated with blocking buffer instead of primary antibody. The ImageExpress MicroXLS High-Content Analysis System using the MX6 Software (Molecular Devices) was used to image slides at 20×.

Results

Microglia scRNA-seq

An average of 3838 ± 553 cells passed quality control metrics per experimental group, with an average of 44,768 ± 6308 reads per group. Cells were predominantly microglia (15,355 total; 96.2–97.7% of cells sequenced, Figure 1A). A pronounced shift was observed in the transcriptional profile of the hippocampal microglia in TgF344-AD rats as compared to wild-type rats (Figure 1B, Supplemental Figure 3). At 3 months of age, 430 transcripts were differentially expressed (p_FDR<0.05) and, by 10 months of age, the number of differentially expressed transcripts doubled to 934 (Supplemental Table 1).

Figure 1.

Successful isolation and scRNA-seq of microglia in the TgF344-AD rat model. (A) Cell type was clearly defined as predominantly microglia. (B) Transcriptional profiles clustered predominantly by age (3 versus 10 months) and then by genotype (AD versus WT), with the greatest distinction between genotypes apparent at 10 months of age. AD: TgF344-AD; WT: wild-type.

At 3 months of age, the top upregulated transcripts in AD rats included the neuroprotective DAM genes ApoE and Trem2. By 10 months of age, the profile had shifted and Trem2 was no longer significantly upregulated. Instead, neurotoxic DAM transcripts, including Gpnmb, Cd74, Cd9, and Bst2 were upregulated in AD animals (pink; Figure 2, Supplemental Figure 4). We also identified significant upregulation of human-specific MHC class-II transcripts in the TgF344-AD rat (purple; Figure 2). Human-specific Irf8 was also significantly upregulated, but at a lower log₂FC (purple; Figure 2). At 10 months, top downregulated transcripts included homeostatic markers Tmem119, Selplg and Ctss (green; Figure 2) and human-specific Aβ-associated transcripts Ifngr1 and Fkbp5^23,24 (purple; Figure 2).

Figure 2.

Differentially expressed transcripts in hippocampal microglia from 9-month-old TgF344-AD rats. Microglial profiles of homeostatic markers (green) and human-specific Aβ-associated transcripts (purple) were significantly downregulated. Classic neuorotoxic DAM markers (pink) and human-specific Aβ-associated transcripts (purple), including Irf8 and MHC class-II genes, were significantly upregulated.

RNA in-situ hybridization

Gene expression of Gpnmb increased with age in both wild-type and transgenic rats (p = 0.0006 and p < 0.0001, respectively, Figure 3 and Supplemental Figure 5). At 3 months of age, there was no difference in expression of Gpnmb between wild-type and transgenic rats (p = 0.32). At 10 months of age, however, a significant increase in Gpnmb expression was identified in TgF344-AD rats (p = 0.0003, Figure 3 and Supplemental Figure 5). While there was no difference in gene expression of Tmem119 at 3 months (p = 0.95) or 10 months (0.37) of age, Tmem119 expression increased with age in wild-type rats (p < 0.0001) but only a trend was observed with aging in transgenic rats (p = 0.07, Supplemental Figure 6). Subsequent immunofluorescence staining confirmed that the ISH signal was from microglia (Supplemental Figure 7). Thus, ISH confirmed the scRNA-seq results for Gpnmb in microglia, but we only identified a trend for decreased Tmem119.

Figure 3.

Differentially expressed Gpnmb in ventral hippocampi of 3 and 10-month-old wild-type (WT) and TgF344-AD (Tg) rats using in situ hybridization. Expression of the DAM marker Gpnmb (red) was significantly increased with age in both genotypes but significantly higher in TgF344-AD rats at 10 months of age. ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05, ns = not significant, 2-way ANOVA with Tukey's post hoc test, n = 4–11 per group. Quantification included all regions of the hippocampus and entorhinal cortex. Representative images of the dorsal hippocampi are in Supplemental Figure 5. To validate that transcripts were within microglia, immunofluorescence staining was performed on the CA3 region immediately adjacent to the ISH images (Supplemental Figure 7). Scale bar=100 μm.

Discussion

We successfully demonstrated a downregulation in homeostatic microglial markers and concurrent increase in first neuroprotective, and then neurotoxic, DAM markers with progression of AD in the TgF344-AD rat model. The microglial signature with human AD has been suggested to be driven by MHC class-II transcripts^7,12 and IRF8,^7,8 which are not consistently identified in mouse DAM.^7,10 Importantly, we identified significant upregulation of MHC class-II transcripts and a significant, albeit less substantial, increase in IRF8 in the TgF344-AD rat. Thus, TgF344-AD rats better recapitulate the microglial gene expression changes observed in human AD.

For confirmation of scRNA-seq results, we selected one highly upregulated transcript, Gpnmb and one highly downregulated transcript, Tmem119. Glycoprotein NMB, which is part of the microglia activation state, has been proposed to be a novel disease-associated marker that plays a role in the neuroinflammatory response of AD. ²⁵ In a mouse model of AD, Gpnmb expression increases in parallel with Aβ plaque deposition, reflecting disease severity. ²⁵ GPNMB + microglia also moderate the Aβ-tau interaction in early AD. ²⁶ Increases in GPNMB expression may actually be beneficial, through enhanced autophagy and promotion of Aβ clearance. ²⁷ Therefore, the TgF344-AD rat is an excellent animal model to further investigate the role and regulation of GPNMB in AD.

Based on transcriptomic findings, TMEM119 was considered to be a marker of microglia under homeostatic conditions in humans and mice.^1,28 However, contrary to these results, increased mRNA levels have been found in brains of some patients with AD. ²⁹ When postmortem AD brain tissue was evaluated using multispectral immunofluorescence, a significant reduction of TMEM119 + microglia was identified, although not associated with Aβ plaques. ²³ Thus, while TMEM119 is not consistently expressed by microglia, it is generally lost in AD. Despite our inability to validate our scRNA-seq results for Tmem1119 with ISH, the TgF344-AD rat may still recapitulate this phenotype, with a trend towards lower Tmem119 in the transgenic rats at 10 months of age.

While in situ hybridization was successful for both Gpnmb and Tmem119 transcripts, our protocol was not compatible with adding an additional marker for Iba1. In order to define microglial populations in the ISH positive region, immunofluorescence staining was performed on the CA3 region immediately adjacent to the ISH images (Supplemental Figure 7). Since sections were sequential, a limitation of this study was that we were unable to directly overlap these images to validate that targeted transcripts were specific to microglia. However, the two different stains were performed on 10 µm thick serial sections and, with an average cell body perimeter of 33 µm (i.e., diameter of 10.5 µm), ²⁴ these were very likely the same microglial population.

TgF344-AD rats are one of the most suitable animal models for AD research, with age-dependent manifestations of cerebral amyloidosis, tauopathy, oligomeric Aβ, gliosis, apoptotic loss of neurons, and behavioral impairment. ¹³ Our data suggests that the TgF344-AD rat also better recapitulates the microglial gene expression changes observed in human AD.

Supplemental Material

sj-xlsx-1-alz-10.1177_13872877251410206 - Supplemental material for Microglial transcriptional profiles of a transgenic rat model closely model Alzheimer's disease

Supplemental material, sj-xlsx-1-alz-10.1177_13872877251410206 for Microglial transcriptional profiles of a transgenic rat model closely model Alzheimer's disease by Carrie J. Finno, Sharmila Ghosh, Veronika Rodriguez, Anthony Valenzuela, Peter Andrew, Heui Hye Park, Ana Cristina Grodzki, Nathifa Nasim, Kelsey Roberts, Blythe Durbin-Johnson, Ken A. Jackson, Patricia A. Pesavento and Pamela J. Lein in Journal of Alzheimer's Disease

Supplemental Material

sj-docx-2-alz-10.1177_13872877251410206 - Supplemental material for Microglial transcriptional profiles of a transgenic rat model closely model Alzheimer's disease

Supplemental material, sj-docx-2-alz-10.1177_13872877251410206 for Microglial transcriptional profiles of a transgenic rat model closely model Alzheimer's disease by Carrie J. Finno, Sharmila Ghosh, Veronika Rodriguez, Anthony Valenzuela, Peter Andrew, Heui Hye Park, Ana Cristina Grodzki, Nathifa Nasim, Kelsey Roberts, Blythe Durbin-Johnson, Ken A. Jackson, Patricia A. Pesavento and Pamela J. Lein in Journal of Alzheimer's Disease