Human Alzheimer’s disease gene expression signatures and immune profile in APP mouse models: a discrete transcriptomic view of Aβ plaque pathology

Male transgenic (Tg) mice expressing APP KM670/671NL (Swe) and APP V717F (TgCRND8) and age-matched wild-type (WT) littermates were used in whole cortex and laser capture RNAseq evaluations as well as in situ hybridization studies. Tg female mice carrying the APPSwe mutation (Tg2576) [33] and age-matched WT littermates were used for the whole-cortex evaluations. Mixed sex cohorts were not available, precluding any evaluation of sex differences in gene expression. Mice were bred for Merck at Taconic Farms (Hudson, NY, USA) and single housed in an Association for Assessment and Accreditation of Laboratory Animal Care accredited facility that was maintained on a 12-h light/dark cycle (lights on at 0700 h). Food and water were available ad libitum. Principles of laboratory animal care were followed, and all studies were previously approved by the Institutional Animal Care and Use Committee and were performed in accordance to the Guide for the Care and Use of Laboratory Animals as adopted and promulgated by the National Institutes of Health (Library of Congress Control Number 2010940400, revised 2011).

For whole cortex analysis of gene expression, TgCRND8 mice were terminally sacrificed using CO₂ at 1.5 (n = 4), 3 (n = 6), 4.5 (n = 3) 6 (n = 5), or 10 months (n = 7) of age. Corresponding WT littermates were also sacrificed at 1.5 (n = 6), 3 (n = 6), 4.5 (n = 4) 6 (n = 5), or 10 months (n = 6) of age. Tg2576 mice and corresponding WT controls were sacrificed similarly at 3, 6, 9, 12, and 15 months of age (n = 4 per group/age). After decapitation, the cortex was removed and rapidly frozen in liquid nitrogen and stored at − 80 until processing. RNA was extracted from TgCRND8 and corresponding control animals using Qiagen RNeasy Mini kit (Qiagen). The Magmax 96 for Microarray RNA extraction kit (Life Technologies) was used to for Tg2576 and corresponding control animals. RNA quality in all samples was assessed with the Agilent2100 Bioanalyzer. Libraries for RNA sequencing were prepared by Beijing Genomics Institute (BGI, Philadelphia, PA) using 100 ng of purified RNA and the Truseq stranded total RNA RiboZero library preparation kit (Illumina) strictly following the Illumina guide (15031048 E?). Briefly, rRNA was depleted with rRNA Removal Mix (RRM) and then fragmented into ~ 160 bp fragments. rRNA-depleted RNA fragments served as templates for first-strand cDNA synthesis using random hexamer-primers, followed by second stand synthesis with the addition of buffer, dNTPs, RNase H, and DNA polymerase I. Double-stranded cDNA was purified using the QiaQuick PCR extraction kit (Qiagen) followed by end repair, base A addition, and ligation of sequencing adapters. Ligated fragments were purified by magnetic beads and amplified via PCR. The resulting library products quantified with the Agilent2100 bioanalyzer and were sequenced using an Illumina HiSeqTM 4000 for a total of 3GB of 50 bp paired-end read data per sample.

NanoString nCounter technology (http://​www.​nanostring.​com/) allows expression analysis of multiple genes from a single sample. We performed nCounter multiplexed target profiling of 400 custom transcripts. The nCounter codesets were selected prior to the RNAseq data; however, significant representation of differentially expressed genes in the whole cortex and LCM datasets were captured by the custom chips. Fifty nanograms of total RNA per sample was used in all described nCounter analyses according to the manufacturer’s suggested protocol. Raw counts for key targets are illustrated in Additional file 1: Figure S1 confirming the RNAseq data via an alternative gene expression profiling platform. Additionally, in Additional file 2: Figure S2, the fold-change data for each transcript on the custom NanoString panel is plotted versus the RNAseq data, for both the whole cortex samples and LCM samples.

Purified RNA (1 ng/sample) from 24 LCM samples (12 transgenic plaque region and 12 transgenic distal non-plaque region) were used for cDNA production using a SMARTer Ultra Low RNA Kit (Clontech, Mountain View, CA) following the manufacturer’s instructions with minor modifications. After cDNA amplification and purification, half of the resulting cDNA was used as input for the Nextera XT DNA Sample Preparation Kit (Illumina, San Diego, CA) using a Nextera XT Index Kit (Illumina). Nextera XT sample preparation uses an engineered transposome to fragment and attach sequencing adapters to DNA samples. Illumina Nextera XT protocols were followed for library preparation with 1.0 volumes of purification Agentcourt Ampure XP beads (Beckman Coulter, Beverly MA). No normalization was performed on individual libraries. The library quality was determined using a NanoDrop 1000 (Thermo Scientific, West Palm Beach, FL) to measure library concentration and Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA) with a DNA 1000 chip to assess library size and quality. Libraries were pooled, and sequencing was performed using an Illumina HiSeq 2500 (Illumina) sequencer with Rapid paired-end sequencing of 76 × 76 bases. Fastq files were generated and demultiplexed using the bcl2fastq version 1.8.4 program.

Alignment and differential gene expression analysis was performed in Omicsoft Array Studio version 7.2.2.29. Briefly, cleaned reads were aligned to the human B37.3 genome reference using the Omicsoft Aligner with a maximum of four allowed mismatches. Gene level counts were determined by the RSEM algorithm as implemented in Omicsoft Array Studio and using RefGene transcript annotation prepared on March 21, 2014. Resultant FPKM data was subsequently ratioed to the average of the WT plaque samples for each gene prior to analysis. Only genes with ≥ 10 counts in 50% of the samples in at least one treatment group where considered detected and included in analysis.

Differentially expressed genes were identified by Pearson correlation or t test using MatLab R2010b (Mathworks). A p value cutoff of < 0.001 was used to identify differentially expressed genes. The FDR corresponding to this p value is given in each of the comparisons to convey relative signature confidence. Set annotation analysis was performed by comparing input sets to GeneGo (www.​genego.​com), Ingenuity (www.​ingenuity.​com), and KEGG (www.​genome.​jp/​kegg/​) pathway sets. Bonferroni-corrected hypergeometric p values (expectation (e) values) of less than 0.1 were considered significant overlap between sets. Heatmaps were generated in MatLab R2010b using agglomerative clustering.

First, we carried out RNAseq on cortexes isolated from Tg2576, TgCRND8, and WT controls. Data revealed a genotype-dependent increase in gene expression differences with respect to age primarily in TgCRND8 animals (Additional file 3: Table S1). In WT animals, relatively few genes showed age-dependent changes. Only 28 genes had Pearson correlation coefficient (|ρ|) > 0.75 with respect to age in 3–15 month Tg2576 littermate controls, while just 112 genes correlated similarly with age in 1.5–10 month TgCRND8 littermate controls. Similarly, 42 genes had |ρ| > 0.75 with age in 3–15 month Tg2576 animals. In contrast, 1994 genes had correlation |ρ| > 0.75 with age in 1.5–10 month TgCRND8 animals. In order to focus on genotype-related differences without confounding normal aging signatures, we compared the expression differences in Tg2576 and TgCRND8 animals to their respective age-matched WT controls. A small number of genes were differentially expressed between Tg2576 and WT at all ages examined (Fig. 1a, b). Specifically, 43, 42, 86, 11, and 58 out of 15,857 detected genes were significantly different (p < 0.001) between WT and Tg2576 animals at 3, 6, 9, 12, and 15 months, respectively (FDR < 37, 38, 18, 100, and 27%, respectively). Combined, 235 genes displayed a significant difference (p < 0.001) between Tg2576 and WT at one or more ages tested, 125 genes (53%) of which were upregulated. However, a strong progression of genotype-related expression changes was observed in TgCRND8 animals (Fig. 1a, b). At 1.5 months of age, 52 genes had significantly different expression in the TgCRND8 cortex compared to WT controls (p < 0.001, FDR < 29%); by 3 months of age, 108 genes (FDR < 14%) were similarly regulated and this number increased to 170 (FDR < 9%) and 351 (FDR < 4%) at 6 and 10 months, respectively. Combined, 582 genes displayed a significant difference (p < 0.001) between TgCRND8 and WT at one or more ages tested, 464 (80%) of which were upregulated. Numerous gene pathways were significantly (Bonferroni-corrected p value (e) < 0.1) enriched among the genes upregulated in TgCRND8 cortex, predominantly those associated with immunity inflammation lipid metabolism and protein salvage (Additional file 4: Table S2). For example, numerous lysosomal and ITIM (immunoreceptor tyrosine-based inhibition motif; S/I/V/LxYxxI/V/L) and ITAMs (immunoreceptor tyrosine-based activation motif; consensus sequence YxxI/Lx6-12YxxI/L) signaling genes exhibited age-dependent increases in expression in TgCRND8 mice (Fig. 1c, d). However, no biological pathways were identified as significantly enriched among the smaller number of genes downregulated in Tg2576 or in either of the upregulated or downregulated gene sets in Tg2576 despite Tg2576 mice presenting with plaque pathology (Additional file 5: Figure S3).

Next, we assessed whether orthologs of the previously identified human cortical gene networks impacted with aging and AD were modulated with age in the mouse models examined (Additional file 6: Table S3) [34]. As expected, little progression of the human aging signatures was observed in mouse cortex across the ages examined (Fig. 1c, d). Although some of the aging signature genes changed significantly, the trends across all genes with age and the differences between Tg2576/TgCRND8 and WT controls were not significant (Fig. 1c, d). The human inflammatory signature was upregulated in aged specimens. This signature was robustly potentiated in AD subjects regardless of age and also exhibited strong progression with TgCRND8 aged animals. Indeed, this signature was highly enriched (p < 3e−53) among the 464 genes differentially upregulated (p < 0.001) between TgCRND8 and WT controls (Fig. 1). Similarly, we examined whether the brain cell-type specific gene sets identified by Zhang et al., 2013 were enriched among the up or down arms of the Tg2576 or TgCRND8 signatures. Significant overlap was only observed with the microglial gene set, which was highly enriched (p < 6e−24) among the genes upregulated in TgCRND8 (Fig. 1c, d) [23].

In order to establish that the identified gene networks, particularly the human-related disease signatures, were not only temporally but also regionally associated with Aβ pathology LCM was performed around Thio-S-labeled sites. We used the in situ hybridization pattern of TREM2 and CD33 around Aβ plaques in the aged TgCRND8 mouse (Fig. 2) to qualitatively define the radius of the laser capture parameters. Representative images were stained with Iba-1 and 6E10 (Fig. 2c, d) to illustrate plaque and non-plaque areas processed for RNA seq. Brain slices prepared from aged TgCRND8 mice and processed for Aβ immunohistochemistry (6E10 labeling, magenta) combined with TREM2 (panel A) or CD33 (panel B) in situ hybridization via RNAScope (brown) confirmed their associated expression pattern with Aβ pathology in the TgCRND8 model and supported our rationale for laser capture microdissection not just of Thio-S-labeled plaques (as performed by [35]) but also the surrounding penumbra (Fig. 2). RNAseq of laser capture samples revealed transcriptomic differences between plaque niche and non-plaque TgCRND8 tissue. In TgCRND8 mice, 82 out of 12,461 detected genes were significantly (p < 0.001; FDR < 15%) regulated compared to non-plaque TgCRND8 tissue; 67 of these genes were upregulated and 15 were downregulated (Fig. 3a, b; Additional file 7: Table S4). Numerous gene pathways were significantly (e < 0.1) enriched among the genes upregulated in the plaque niche, predominantly those associated with vesicles, immunity, inflammation, and protein salvage, for example lysosomal or immunoreceptor tyrosine-based activation/inhibition motif (ITIM/ITAM) signaling genes (Fig. 3c, Additional file 4: Table S2). As in the TgCRND8 versus WT comparison, the human inflammation signature (p < 3e-34) as well as the microglial-specific signature (p < 7e-7) were significantly enriched among the genes upregulated within the plaque niche signature, while the other human and cell-type gene sets were not (Fig. 3c). The expression differences in TgCRND8 mice and those enriched in the plaque niche were highly overlapping. The majority of the genes upregulated in TgCRND8 cortex compared to WT animals were also expressed higher in CRND8 plaque versus non-plaque (Fig. 3a). Conversely, the majority of the genes upregulated in the plaque versus non-plaque comparison were also upregulated in 6–10-month-old TgCRND8 animals compare to WT (Fig. 4b). Several genes (12,130) were detected in both experiments, including 342 genes that were upregulated in TgCRND8 vs WT cortex (p < 0.001 at any age) and 67 genes upregulated in plaque vs non-plaque (p < 0.001), of which 46 were overlapping between both signature sets (p < 2e-56) (Fig. 4c). We confirmed the findings from our RNAseq data using a customized NanoString nChip showing that key plaque-associated genes were upregulated (Trem2, Tyrobp, Cd68, Clec7a, Tspo, Itgfax) (Additional file 1: Figure S1). Data were analyzed for expression of a set of markers associated with resident microglia or peripheral macrophages as outlined by Hickman et al. to characterize the phenotype of cells surrounding the plaque [36]. Results show that 7 out of the top 25 most abundant genes in resident microglia were significantly upregulated in the plaque niche compared to non-plaque samples (Additional file 8: Figure S4). Conversely, only 1 gene associated with peripheral macrophage expression was significantly upregulated in the plaque niche, Complement C4-b (C4b) (Additional file 8: Figure S4). This phenotypic analysis was more pronounced in whole cortex samples resulting in 13 out of the top 25 microglial-specific genes were significantly upregulated in TgCRND8 mice compared to WT compared to only 1 gene in the peripheral macrophage signature (Additional file 8: Figure S4).

To further characterize the microglia phenotype observed in the TgCRND8 plaque niche, we compared the gene signatures identified here to those of the microglial subtype, disease-associated microglia (DAM), recently shown to surround plaques in another mouse AD model expressing five familiar AD mutations (5XFAD) [37]. Indeed, genes reported to be expressed higher in the DAM population (> 1.5 fold, p < 1e-5) were strongly enriched among the upregulated gene signatures in both the TgCRND8 whole cortex and the LCM-dissected TgCRND8 plaque tissue signatures (p = 4e-25 and p = 8e-13, respectively) (Fig. 5a–d; Additional file 4: Table S2, Additional file 6: Table S3, and Additional file 7: Table S4). Across all three gene sets, eight genes were commonly expressed: Ccl6, Cd9, Ctsz, Gusb, Lyr2, Npc2, Trem2, and Tyrobp. In contrast, the DAM marker genes were not significantly represented among the Tg2576-related signatures (p = 0.3) (Fig. 5a, c). In additional analyses, expression of markers of M1 and M2 microglia as outlined in [38, 39] were also analyzed for differential expression in plaque vs. non-plaque samples as well as in whole cortex samples from TgCRND8 compared to WT controls. Results show a significant increase in expression of only one marker of M1 activation, CD16a, in the plaque niche compared to non-plaque controls (3.06-fold; p < 0.0001). No markers of M2 activation were significantly upregulated in plaque samples compared to non-plaque. Whole cortex TgCRND8 displayed significant (p < 0.001) expression of four markers of M1 activation, CXCL10 (34.52-fold; p < 0.0001), CCR5 (1.51-fold; p < 0.0001), CD86 (2.25-fold; p < 0.0001), and CD16a (2.62-fold; p < 0.0001) in TgCNRD8 compared to WT, and only two markers of M2 activation/repair, Clec7a (72.33-fold; p < 0.0001), and TGFβ (2.23-fold; p < 0.0001).