All experimental procedures involving animals in this study were approved by and complied with the guidelines of the Institutional Animal Care and Use Committee of the Nathan Kline Institute and Columbia University Medical Center. Human APOE targeted replacement mice were originally developed by Sullivan et al. (1997, 2004). These targeted-replacement mice express human APOE under the control of the endogenous murine promoter, which allows for the expression of human APOE at physiologically regulated levels in the same temporal and spatial pattern as endogenous murine Apoe. For our initial transcriptomics analysis, 14–15 month old male APOE3/4 vs. APOE3/3 mice were used, whereas immunohistochemical analysis of endosomal and lysosomal morphology was performed on approximately equal numbers of male and female APOE4/4 vs. APOE3/3 mice at 6, 12, 18, and 25 months of age. Mice were derived from two separate colonies. The APOE genotype of the mice was confirmed either in-house by restriction fragment length polymorphism analysis following polymerase chain reaction (PCR), as previously described (Hixson and Vernier, 1990) or using a PCR-based method developed by Transnetyx (Cordova, TN).

Mouse numbers were selected to be similar to those used in previous experiments (Hauser et al., 2014; Dillman et al., 2016). Male mice expressing human APOE3/3 (10 mice) or APOE3/4 (19 mice) were aged to 14–15 months, at which point they were sacrificed by cervical dislocation, and brain tissues containing the EC and PVC were dissected and snap frozen on dry-ice. Brain tissues were stored in RNase-free eppendorf tubes at −80°C prior to extraction. Total RNA was extracted from frozen tissues by homogenizing each tissue sample using a battery-operated pestle mixer (Argos Technologies, Vernon Hills, IL) in 1 ml of TRIzol reagent according to the manufacturer's protocol (Life Technologies, Carlsbad, CA). RNA concentration was measured using a nanodrop 1000 (Thermo Fisher Scientific, Waltham, MA), and RNA integrity (RIN) was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). All RNA samples possessed a RIN of 8 or higher. RNA was stored at −80°C prior to use.

Starting with 2 μg of total RNA per sample, Poly(A)+ mRNA was purified, fragmented and then converted into cDNA using the TruSeq RNA Sample Prep Kit v2 (Illumina cat# RS-122-2001) as per the manufacturer's protocol (Illumina, San Diego, CA). For RNA-Sequencing of the cDNA, we hybridized 5 pM of each library to a flow cell, with a single lane for each sample, and we used an Illumina cluster station for cluster generation. We then generated 149 bp single end sequences using an Illumina HiSeq 2000 sequencer. For analysis, we used the standard Illumina pipeline with default options to analyze the images and extract base calls in order to generate fastq files. We then aligned the fastq files to the mm9 mouse reference genome using Tophat (v2.0.6) and Bowtie (v2.0.2.0). In order to annotate and quantify the reads to specific genes, we used the Python module HT-SEQ with a modified NCBIM37.61 (containing only protein coding genes) gtf to provide reference boundaries. We used the R/Bioconductor package DESeq2 (v1.10.1) for comparison of aligned reads across the samples. We conducted a variance stabilizing transformation on the aligned and aggregated counts, and then the Poisson distributions of normalized counts for each transcript were compared across APOE3/4 vs. APOE3/3 groups using a negative binomial test. We corrected for multiple testing using the Benjamini-Hochberg procedure and selected all genes that possessed a corrected p-value of less than 0.05. Finally, a heat map based on sample distance and a volcano plot based on fold change and adjusted p-values were generated using the R/Bioconductor package ggplot2 (v2.0.0).

Enriched KEGG pathways were identified using the ClueGo application (version 2.1.7) in Cytoscape (version 3.2.1). Briefly, all differentially expressed EC genes from the RNA-Seq analysis were entered into the application and searched for significantly enriched KEGG pathways possessing a Benjamini-Hochberg adjusted p-value of less than 0.05. For gene ontology (GO) pathway analysis, differentially expressed EC genes were entered into NIH's DAVID program (https://david.ncifcrf.gov/; version 6.8), and a functional annotation chart was generated for all significantly enriched GO Biological Process Terms, GO Cellular Component Terms and GO Molecular Function Terms possessing a Benjamini-Hochberg adjusted p-value of less than 0.05.

For immunolabeling of brain tissue, mice were transcardially perfusion-fixed with 4% paraformaldehyde in phosphate buffered saline (PBS). Brains were removed and post-fixed for 3 days in 4% paraformaldehyde in PBS at 4°C, and subsequently cut into 40 μm-thick coronal sections using a vibratome, as we have described previously (Choi et al., 2013). Free-floating sections were labeled with anti-Rab5a (sc-309; Santa Cruz Biotechnology, Santa Cruz, CA; dilution 1: 100) or an in-house anti-CatD rabbit polyclonal antibody (Yang et al., 2014). Tissue sections were also co-labeled with a Nissl stain to determine the border of the cell soma. Following binding of appropriate fluorophore-conjugated secondary antibodies (Invitrogen, Grand Island, NY; dilution 1:500), immunofluorescence was captured at 100 × magnification using an Axiovert 200M Fluorescence Imaging Microscope (Zeiss, Thornwood, NY) by a genotype-blinded observer. Quantification and morphometric analysis of labeled particles per cell was measured using ImageJ's automated particle analysis (NIH, Bethesda, Maryland) after thresholding the density of Rab5a or CatD-positive signal over background, as we have done previously (Choi et al., 2013). With this method, we obtained automated calculations of average area, particle count, and area fraction within a manually drawn border around the cell soma. Statistical analysis was performed using a Student's t-test between groups. Cell and mouse numbers were selected to be similar to those used in previous experiments (Choi et al., 2013; Jiang et al., 2016). For the quantifications in EC, we used n = 118 neurons from 7 mice (APOE3/3) and n = 85 neurons from 6 mice (APOE4/4) for Rab5a, and n = 126 neurons from 7 mice (APOE3/3) and n = 139 neurons from 7 mice (APOE4/4) for CatD. For the quantifications in cingulate cortex, we used n = 48 neurons (APOE3/3) and 52 neurons (APOE4/4) from 4 mice per genotype at age 6 months, n = 115 neurons (APOE3/3) and 88 neurons (APOE4/4) from 8 mice per genotype at age 12 months, n = 267 neurons (APOE3/3) and 286 neurons (APOE4/4) from 8 mice per genotype at age 18 months, and n = 342 neurons (APOE3/3) and 141 neurons (APOE4/4) from 8 mice per genotype at age 25 months.

The transcriptomics datasets generated during the current study are available in the NCBI Gene Expression Omnibus (GEO) repository, accession # GSE102334.

In order to investigate the effect of APOE4 expression in an AD-vulnerable vs. an AD-resistant brain region, RNA-sequencing was performed on RNA extracted from the EC and PVC of 14–15 month-old APOE targeted replacement mice expressing human APOE3/4 (19 males) vs. APOE3/3 (10 males). As shown in Figure 1A, the samples clustered independently based on brain region, but not based on genotype. This is likely due to the relatively small number of genes that were differentially expressed between APOE3/4 and APOE3/3 mice in each region versus the total number of genes analyzed. Using an FDR-corrected p-value of less than 0.05, we identified 1292 genes in the EC and 55 genes in the PVC that were differentially expressed between APOE3/4 and APOE3/3 (Tables S1, S2), out of a total of 31,246 genes analyzed. In the EC, 683 genes were decreased and 609 genes were increased in APOE3/4 mice as compared to APOE3/3 mice, while in the PVC, 28 genes were decreased and 27 genes were increased in APOE3/4 mice as compared to APOE3/3 mice (Figure 1B).

In order to identify specific cellular pathways that are affected by differential APOE isoform expression in the EC, we performed KEGG pathway analysis on the differentially expressed genes from the EC of APOE3/4 vs. APOE3/3 mice using Cytoscape's ClueGo application. This analysis identified 9 KEGG pathways that were enriched with genes from the EC of APOE3/4 vs. APOE3/3 mice (Figure 2 and Table 1). An informal analysis of these results revealed that there were four major pathway groups affected by the APOE3/4 genotype in the EC: vesicle function (synaptic vesicle cycle, collecting duct acid secretion and phagosome pathways), developmental genes (Hippo and Wnt signaling pathways), oxidative phosphorylation (oxidative phosphorylation pathway) and AD (Alzheimer's disease pathway).

While all of these KEGG pathway groups are intriguing, we focused on the vesicle function group. As highlighted in Figure 2, an important contributor to the difference in the vesicle function group was an increase in the expression of 9 Vacuolar H⁺-ATPases (V-ATPases) in the EC of APOE3/4 mice (Figure 3A). V-ATPases are vital for the acidification and function of a wide range of vesicular compartments, including phagosomes in macrophages and microglia, synaptic vesicles in neurons, and endosomes, lysosomes and autophagosomes in numerous cell types (Marshansky et al., 2014; Cotter et al., 2015). In order to further explore this pathway enrichment, we also performed a DAVID analysis for Biological Process (BP), Cellular Component (CC) and Molecular Function (MF) Gene Ontology (GO) terms (Table 2), and we found that differentially expressed EC genes were enriched in “endosome,” “lysosomal membrane” and “extracellular exosome” GO terms, in addition to numerous other vesicle and transport-related GO terms. For this reason, we chose to direct our attention to the effect of differential APOE isoform expression on the endosomal–lysosomal system.

In addition to the increased expression of V-ATPases, further mining of the RNA-Seq data revealed 12 different Rab GTPases and associated genes (Figure 3B), including Rab5b and Rab7, which are vital for the formation and function of early and late endosomes, respectively, as well as several other important regulators of endosomal–lysosomal function, including Atp6ap1, Atp6ap2, Snx3, Snx15, Vps4a, Vps24, Vps29, Vta1, and Mp6r. Intriguingly, all of these genes were found to be upregulated in the APOE3/4 EC, which we hypothesize might represent a compensatory effect in response to a broad dysregulation of the endosomal–lysosomal system in the EC of these APOE3/4 mice. Importantly, none of these genes were shown to be differentially expressed in the PVC, suggesting that APOE4-mediated endosomal–lysosomal pathology may be occurring predominantly in AD-vulnerable brain regions such as the EC.

Abnormalities in the endosomal–lysosomal system are prominent in AD (see review by Nixon, 2017). Enlarged early endosomes are among the first cellular pathological features observed in AD pathogenesis (Cataldo et al., 2000, 2004; Nixon et al., 2000; Nixon, 2005), even prior to amyloid deposits in a given brain region, and AD-related enlarged early endosomes occur to a greater degree in the brains of APOE4 carriers (Cataldo et al., 2000). Building upon our transcriptomics findings and to determine whether endosomes are similarly affected by APOE4 expression in the absence of overt AD pathology, we investigated the endosomal–lysosomal system in the brains of 18-month-old APOE4/4 vs. APOE3/3 targeted replacement mice, as has been done in human tissue (Cataldo et al., 2000). 18-month-old APOE4/4 vs. APOE3/3 mice were chosen in order to gain a more robust picture of the endosomal–lysosomal changes observed in the transcriptomics results. Focusing initially on the EC of these mice, we quantified Rab5a immunolabeling of early endosomes in pyramidal neurons (Figure 4), as we have done previously on other AD-related mouse models (Choi et al., 2013; Jiang et al., 2016). In APOE4/4 mice, we observed an increase in the mean number (8.5 ± 0.5 APOE3/3, 11.7 ± 0.8 APOE4/4 particles/cell; p = 0.0007; Figure 4B) and area fraction (0.6 ± 0.05% APOE3/3, 0.8 ± 0.07% APOE4/4 of cell area; p = 0.02; Figure 4D) of Rab5a-positive neuronal early endosomes when compared with age-matched APOE3 mice. No significant differences were seen in average endosomal size between the genotypes (140 ± 7.0 APOE3/3, 157 ± 8.0 APOE4/4 in nm2; p = 0.12; Figure 4C).

To analyze the effect of APOE4 genotype on distal endosomal–lysosomal pathway compartments in neurons, we performed immunolabeling for cathepsin D (CatD), an acid protease that is localized primarily to lysosomes (Figure 4). As with the Rab5a-positive early endosomes, we saw an increase in the mean number (7.5 ± 0.3 APOE3/3, 8.6 ± 0.4 APOE4/4; p = 0.04; Figure 4B) and area fraction (1.2 ± 0.8 APOE3/3, 1.6 ± 0.11 APOE4/4; p = 0.02; Figure 4D) of CatD-positive compartments in the EC of APOE4/4 mice, as compared to APOE3/3 controls. Average size of the CatD-positive compartments was not changed between genotypes (288.5 ± 15.8 APOE3/3, 317.1 ± 22.3 APOE4/4; p = 0.22; Figure 4C).

In order to further characterize the effect of APOE4 on early endosomes, we performed a larger study of Rab5a-positive early endosomes in the cingulate cortex of 6, 12, 18, and 25 month old APOE4/4 vs. APOE3/3 mice (Figure 5). The cingulate cortex was chosen because, like the EC, it is vulnerable to AD pathology (Leech and Sharp, 2014) and to APOE4-specific deficits (Valla et al., 2010). Additionally, we previously characterized early endosomal alterations in this region in amyloid precursor protein transgenic mice (Choi et al., 2013) and, as it is larger than the EC, we anticipated that characterizing early endosomes in this region would facilitate our aging analysis. While no differences were seen in early endosome size, number or area fraction at 6 or 12 months of age, an increase in mean early endosome number (an increase of 30.5 ± 4.0%; Figure 5B), size (33.5 ± 8.0%; Figure 5C), and area fraction (70.1 ± 4.8%; Figure 5D) was seen in the 18 month-old APOE4/4 mice as compared to APOE3/3 controls, a difference that persisted in the 25 month-old animals (endosome number: 73.3 ± 8.4%; size: 20.7 ± 2.7%; area fraction: 71.8 ± 10.4%). This magnitude of early endosome change is similar to that which was reported in Down syndrome mouse models (Cataldo et al., 2003), and is similar to the APOE4-dependent early endosomal alteration seen in human neurons in the prefrontal cortex in early-stage AD (Cataldo et al., 2000).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.