The term “dementia” refers to multiple etiologies and is therefore difficult to define. The simplest definition for dementia therefore can be a progressive loss of brain function, characterized by memory, behavior, and communication problems. Loss of brain function translates into loss of neuronal function; therefore a true animal model of dementia must exhibit signs of progressive ­neuronal dysfunction. There are no published reports of apolipoprotein E (APOE)-based models that exhibit progressive neuronal ­dysfunction. There are, however, several reports of APOE models that portray indirect measures of neuronal dysfunction or ones that might be characterized as exhibiting pre-dementia or mild cognitive impairment (MCI). The following review summarizes the attempts to classify various APOE animal models of “pre-dementia.”

The first evidence to demonstrate an association between the APOE gene and dementia was genetic linkage analysis performed at Duke University in 1993, where the APOE locus on human chromosome 19 repeatedly showed a high correlation with both disease onset and frequency of Alzheimer’s disease (AD). AD is the most common form of dementia. It was later shown that the common APOE genetic polymorphisms were linked to 90% of cases <65 years of age and 60% of AD cases >65 years (1). The human APOE gene encodes one of three alleles designated as APOE*2, E*3, or E*4, which in the Caucasian population occurs at frequencies of 7.3%, 78.3%, and 14.3%, respectively (2). The presence of an APOE*4 allele decreases the average age of onset and increases the risk of AD (3, 4). The APOE*2 allele lowers the risk of AD and increases the average age of onset compared to the APOE*3 allele (5). Each APOE isoform differs by a single amino acid at positions 112 and 158, which results in significant structural and functional differences (6). Since the early 1970s, APOE had been studied extensively because of its relationship to heart disease, and also because its function as a lipid metabolism protein became well established. Today, multiple theories abound to explain the function of APOE in the brain. The APOE made in the brain (primarily of glial origin) is separate from APOE made in the periphery (primarily of hepatocellular origin). ApoE is secreted as a 34 KDa glycoprotein, where it is thought to exist as the major apolipoprotein and lipid carrier in the brain (7). Brain APOE is secreted primarily by astrocytes, and it resides on high-density lipoprotein (HDL)-like particles. Due to its amphipathic nature it is drawn to lipids while still retaining the ability to bind receptors at the cell surface (e.g., neuronal APOE receptors). In vitro studies show that each isoform exhibits significant differences in affinity for the low-density lipoprotein (LDL) receptor (8, 9), however, these studies were performed with nonneuronal cell types. The difficulty of performing in vivo APOE–receptor studies in the brain limits our understanding of this interaction. Furthermore, each isoform has been shown to exhibit significant difference in protein stability, redox status, immune activation, and other purported functions, underscoring the complexity of APOE biology. ApoE-based animal models at minimum provide us with the opportunity to test cell-based theories in a system-wide approach where all the necessary parts are able to interact in a human-like fashion.

The APOE-deficient (–/–) mouse was the first animal used to study APOE’s role in dementia (or more specifically AD). The first studies in APOE (–/–) mice showed age-dependent deficits in ­specific neuronal markers such as synaptophysin and microtubule-associated-2 protein (10). Other studies showed reduced choline acetyltransferase activity (11), and long-term potentiation (LTP) deficits in APOE (–/–) mice (12) compared to wild-type mice. The APOE (–/–) mice also exhibited behavioral deficits in spatial memory that could be rescued by infusion of recombinant APOE3 or E4 (10). Multiple studies were performed with these animals (for review see (10) and (13)) demonstrating cognitive deficits similar to human “pre-dementia.” However, other groups were unable to replicate these findings, casting doubt over the utility of these animals for future dementia studies (14, 15). It has been suggested that differences in genetic background or environment might explain the discrepancy observed in varying studies of APOE (–/–) mice. It became apparent that refining our understanding of APOE’s relationship with dementia would require transgenic mice expressing all three isoforms.

The majority of disease-linked genes involve a mutated (disease causation) and a normal gene. The APOE polymorphism is unique in that one allele (E*4) is linked to disease, another (E*3) is neutral, and the least common allele (E*2) is considered protective. Therefore, a good animal model of APOE-based dementia should include all three alleles as it is equally important to determine why E*4 is harmful as why E*2 is protective (with E*3 serving as the control). This led to the creation of APOE transgenic mice expressing all or a subset of the APOE alleles via standard pronuclear injection of human APOE DNA. The first attempt was performed by Bowman et al. wherein human APOE cDNAs driven by the transferrin (TF) promoter were injected into wild-type mice (not crossed to APOE (–/–) mice) and analyzed for brain APOE expression. Very little characterization of these lines was performed, other than to show the mice expressed brain APOE RNA and APOE3 protein (16). A more extensive study was performed using human APOE transgenic mice created by Xu et al. These mice were created via insertion of human APOE genomic fragments driven by 3–15 kb of upstream regulatory sequences for AD modeling studies (17). Like most transgenic animals, APOE expression was variable in both the amount of APOE protein produced and the spatial distribution pattern as dictated by each founder line. Human APOE could be detected in a subset of neurons by immunocytochemistry, which varied by APOE genotype (founder-specific) and was detected sparingly in astrocytes (18). Today we know that brain APOE is synthesized predominantly in glial cells (i.e., astrocytes) and can only be detected in neurons post-injury. Even in the post-injury state, neuronal APOE expression is difficult to detect, and its importance in AD pathogenesis remains to be determined. In an attempt to answer this question, transgenic mice containing the neuron-specific enolase (NSE), an upstream promoter of a human APOE cDNA, was created (19). These mice expressed APOE at extraordinarily high levels in neurons (only), and exhibited behavioral deficits (19) and an inability to protect against kainic acid-induced injury (20). To avoid confounding effects from mouse APOE expression, these mice were crossed to an APOE (–/–) mouse, which adds a different layer of complexity to the overall phenotype. The NSE–APOE/APOE (–/–) doubly transgenic mice do not express APOE in any glial cells or peripheral cells (e.g., hepatocytes and macrophages). Therefore, these mice have the same phenotype of APOE (–/–) mice, which is hypercholesterolemia and chronic inflammation. This along with an extremely high level of neuronal APOE expression makes interpretation of the resultant phenotype difficult. Other attempts to address the APOE neuronal expression theory used promoters such as platelet-derived growth factor (PDGF), thy1, and phosphoglycerate kinase (PGK) to derive expression of human APOE4 in a wild-type mouse APOE background (21). Mice that had the highest level of neuronal APOE4 expression coupled with advanced age showed evidence for increases in hyperphosphorylated tau (22). No comparison to similar mice expressing APOE3 or E2 was published. A different group did look at APOE2 expression (driven by PDGF or TF) in mice with an APOE (–/–) background showing that APOE2 expression restored levels of synaptophysin that were reduced in APOE (–/–) mice (23).

In an attempt to create an animal that more closely matches the “normal” pattern of brain APOE expression, glial fibrillary acidic protein (GFAP) transgenic mice were made. These mice express APOE only in astrocytes, and because they are crossed with APOE (–/–) mice they lack peripheral expression of APOE (24). There are no overt differences between GFAP-E3 or E4 mice and no brain pathology suggestive of dementia in these mice. However, the GFAP-E4 mice do show reduced spine density compared to the GFAP-E3 mice (25). The GFAP mice suffer the same limitations as other transgenic mice that must be crossed on to an APOE (–/–) background (see above).

At the time, use of transgenic mice for studying human APOE expression provided us with valuable clues to the function of brain APOE. We also learned that transgenic technology produced ­several caveats. This method typically produces mice with varying levels of APOE expression due to differences in chromosomal ­location and copy number of the transgene. The use of artificial promoters also results in imprecise spatial and temporal expression of the transgene in both brain and other tissues where APOE is normally expressed (e.g., liver, kidney, gonads, adrenals). Since all transgenic animals must be crossed to an APOE (–/–) animal to remove the influence of mouse APOE, the peripheral cholesterol levels (∼600 mg/dl) and the subsequent systemic inflammation can lead to unknown isoform-specific consequences in the central nervous system (CNS). Phenotypic differences due to transgene insertional mutagenesis cannot be ruled out, and expression of unwanted sequences within the transgenic construct can complicate the interpretation of the data. Therefore, it is important when designing experiments to investigate APOE isoform-specific effects, that a model with fewer confounding factors should be used.

One way to avoid confounding effects from an APOE (–/–) background is to create APOE “knock-in” mice. Hamanaka et al. (26) created APOE4 “knock-in” mice by targeting the human APOE4 cDNA to the mouse APOE locus. Unfortunately, this strategy resulted in APOE4 mice exhibiting a hypercholesterolemic pheno­type (i.e., plasma total cholesterol was 2.5-fold higher than wild-type mouse controls). This may have been due to the lack of intronic enhancers and/or genomic placement of the neomycin-resistant gene. To address this concern, a separate group bred these mice to ZP3-cre mice to remove the neomycin gene, which effectively eliminated the hypercholesterolemia phenotype from the APOE4 mice. No overt differences between APOE3 and E4 mice in this group have been identified (27).

Possibly the simplest and least complicated APOE animal model available today are the human APOE targeted replacement (TR) mice. In contrast to the “Hamanaka et al.” strategy, these mice were created by replacing the mouse APOE gene with a human APOE genomic fragment (28–30). These animals express human APOE at physiological levels in both a temporal and spatial pattern similar to wild-type mice, nonhuman primates, and humans (31). This similarity in expression pattern across species is due to the fact that the human APOE gene is driven by endogenous mouse regulatory sequences, thus keeping the biofeedback loop intact for studying modulation by extrinsic factors (e.g., diet, oxidative stress, and injury). Although this model has been ­criticized (32) for not including human regulatory sequences, there are no animal models that contain sufficiently large segments (e.g., 100–200 kb) of the human APOE gene for gene expression studies. The TR mice are currently our best alternative for modeling human APOE expression in vivo. Furthermore, the APOE3 and E4 TR mice are normolipidemic when maintained on a normal rodent chow diet and exhibit no overt pathology in any tissue examined. Thus, like human APOE4 carriers, other genetic and nongenetic factors are required to precipitate disease. Most importantly, all three APOE lines (E2, E3, and E4) were made in identical fashion and thus can be used for APOE isoform-specific comparison studies.