Alzheimer’s Disease (AD) is the sixth leading cause of death in the United States. There are two forms of AD: early-onset or familial AD (EOAD), which develops before age 65, and late-onset AD (LOAD). EOAD is caused by autosomal dominant mutations in 3 genes – Amyloid Precursor Protein (APP), Presenilin 1 (PS1), and Presenilin 2 (PS2). The LOAD develops later in life, and in some individuals above 85, with no causative gene mutations known. LOAD cases account for more than 95% of all AD cases. An estimated 6.5 million Americans aged 65 and older were living with Alzheimer’s in 2022. Seventy-three percent are age 75 or older. The cost of Alzheimer’s and other dementias (ADOD) to the nation in 2022 was calculated at $321 billion, and by 2050, these costs could reach nearly $1 trillion. More than 11 million Americans provide unpaid care for people with Alzheimer’s or other dementias. In 2021, these caregivers provided more than 16 billion hours of care valued at nearly $272 billion. The financial burden on American society from ADOD is enormous.

In addition to the cognitive decline, there are two morphological hallmarks of AD: extracellular deposits of β-amyloid (Aβ) peptide, called amyloid plaques and intracellular neurofibrillary tangles of tau protein (Jack et al., 2018; Chen and Holtzman, 2022). While it has been established for more than 30 years now that the highest risk for LOAD is associated with a specific allele of APOE gene – APOEε4, other common gene variants have been added to a long and ever-increasing list of genetic risk factors of various significance (Pimenova et al., 2017; Sims et al., 2017; Ando et al., 2022; Holstege et al., 2022). Half of those are associated with immune response (Wes et al., 2016). Among those, rare variants of TREM2, expressed in microglia, are associated with a risk close to a risk associated with the inheritance of a single APOEε4 allele (Wolfe et al., 2018b). Environmental exposures, lifestyle, diet, traumatic brain injury, and an array of comorbidities have been implicated in LOAD risk, early pathogenesis, and progression, as well (Rao et al., 2023). Unfortunately, none of the knowledge regarding the risk factors of LOAD has translated into early diagnosis of the disease or meaningful direction toward successful therapeutic strategies. In this review, we discuss genetic and epigenetic regulatory mechanisms controlling and influencing APOE expression and focus on interactions of APOE and TREM2 in the context of microglia and astrocytes’ role in AD-like pathology in animal models.

APOE an extremely important and indispensable protein expressed in multiple tissues and organs. It is important to underline that no pathological condition or disease presents with the lack of APOE due to genomic deletion. Functionally, at a biochemical level, APOE provides a scaffold for and is an integral structural part of lipoproteins. In brain, APOE is secreted primarily by astrocytes and, unlike in the periphery, is the major apolipoprotein of High Density Lipoprotein (HDL)-like discoidal particles. These brain HDL-like particles do not contain APOA-I. Thus, the transport of cholesterol and phospholipids in the interstitial fluid and between neural cells highly depends on APOE. The transport of cholesterol and phospholipids is the major function of APOE. Therefore, in case of presumptive dysfunctional APOE, it is reasonable to expect, as a consequence, multiple disturbed molecular and cellular processes. The term dysfunctional APOE is poorly defined, however (Fazio et al., 1994). In AD nonmutated APOE, regardless of the isoform, retains its normal biochemical function.

In an excellent review published 23 years ago, R. Mahley and S. Rall concluded that “APOE plays a part in many processes beyond its traditional role in cholesterol and lipoprotein metabolism” (Mahley and Rall, 2000). A limitation of understanding APOE as a risk of AD is that it is impossible to disentangle the “traditional”/biochemical role of APOE in cholesterol metabolism and its role in AD risk and pathology. Human APOE and its ε2/ε3/ε4 alleles have been associated with many diseases and pathological conditions, including nonpathological aging, and AD (Corder et al., 1993, 1994; Roses et al., 1994; Davies et al., 2014; Table 1). This supports the statement that the genetic risk conferred by APOE is related to a perturbed primary function of APOE isoforms: cholesterol and phospholipid transport and metabolism. This, inevitably, includes AD. Of course, it would have been naïve to explain all of the above phenotypes, particularly AD, only by the disturbed major biological effect of APOE – cholesterol and phospholipid transport.

Numerous proposed hypotheses explain how APOE isoform-specific differences might increase the risk of AD, ranging from neurotoxic effect based on domain interaction, binding to, deposition and clearance of Aβ, differential lipidation of isoforms, and neurotoxic and neuroprotection effects. More than 30 years since the association of APOEε4 and the risk of AD has been established (Corder et al., 1993), we still do not know how exactly the role of APOE coded by APOEε4 allele (APOE4) in cholesterol and lipoprotein metabolism – normal or disturbed, translates into an increased risk of AD. Nevertheless, in the last 2 years, several reviews and research papers have been published pointing to aspects of APOE biology that might be worth considering in terms of deciphering the role of APOE as a risk factor of AD and even designing new therapeutic strategies (Parhizkar et al., 2019; Chen and Holtzman, 2022; Li et al., 2022; Lindner et al., 2022; Martens et al., 2022; Raulin et al., 2022). In an in-vitro experimental system, Lindner et al. (2022) explored isoform-specific lipidation and revealed different lipidation pathways. While ATP Binding Cassette transporter A1 (ABCA1)-regulated APOE lipidation is isoform independent and cholesterol-rich HDL-like lipid particles are secreted by astrocytes, in stress-associated conditions, assembling and secretion of triacylglycerol-rich lipoproteins is boosted by the APOE4 isoform. The authors showed that APOE4 was a strong triacylglycerol binder and thus had a reduced capacity to clear toxic fatty acids from the extracellular milieu. Since 2003 (Koldamova et al., 2003), the importance of Nuclear Receptor’s Liver X Receptors/Retinoid X receptors/ABCA1-APOA-I/APOE (LXR/RXR-ABCA1-APOA-I/APOE) regulatory axis for normal/physiological function of APOE, and its relevance to AD pathogenesis in particular, has been demonstrated in tens of studies [for comprehensive reviews see Fitz et al. (2019), Koldamova and Lefterov (2007), Pahnke et al. (2021), Wolfe et al. (2018a), and Kim et al. (2009)]. The regulatory axis does not simply imply transcriptional regulation. While ABCA1 is a primary LXR/RXR target gene, APOE and APOA-I are part of the axis because transcriptional upregulation of ABCA1 and ABCA1-mediated lipidation of APOA-I and APOE are prerequisites for their stability, avoiding fast apolipoprotein degradation.

Perhaps the most comprehensive review of disturbed molecular, cellular, and pathophysiological processes in AD that might be associated with APOE4 has been published by researchers previously or currently affiliated with Mayo Clinic and Washington University at St Louis (Martens et al., 2022). Alike to Amyloid cascade hypothesis proposed in the late 90s [see J. Hardy for a comprehensive review and critical reappraisal (Hardy, 2009)], the authors coined the term “APOE cascade hypothesis” in the pathogenesis of AD and related dementias. Besides the excellent description of disturbed molecular and cellular processes during AD progression, Martens et al. propose a targeted engagement type of therapeutic strategy based on APOE biology and the physical properties of multimolecular complexes where APOE is an integral part. To be successful, these strategies should nevertheless consider that APOE is not a cause of AD as well as the role of APOE isoforms in development and normal CNS physiology.

Molecular clues to understand the regulated expression of Apolipoprotein E/Apolipoprotein C-I/Apolipoprotein C-IV/Apolipoprotein C-II (E/C-I/C-IV/C-II) gene cluster came from a series of reports based on studies conducted in J. Taylor’s and D. Mangelsdorf’s laboratories in the late 90s and the beginning of this century (Simonet et al., 1993; Allan et al., 1995, 1997; Shih et al., 2000; Grehan et al., 2001; Laffitte et al., 2001, 2003; Mak et al., 2002). The characterization of transgenic mice overexpressing individual genes of the cluster revealed diverse functions. However, the expression of all members of this apolipoprotein gene cluster is reported to be coordinately regulated by distal enhancer regions (Simonet et al., 1993; Allan et al., 1995). It was demonstrated that the regulatory sequences in the MultiEnhancers ME1 and ME2 downstream from the Apolipoprotein C-I (APOC-I) gene are transcription factors Liver X Receptor/Retinoid X receptor (LXR/RXR) Nuclear Receptor dimers canonical response elements (Mak et al., 2002; Figure 1). The regulatory sequences in the promoter regions of the human APOE gene cluster represent binding sites/response elements of additional transcription factors – APP-2, SP1, Estrogen Receptor, as well as transcriptional elements A, B, B1, and B2 (Smith et al., 1988; Jo et al., 1995; Paik et al., 1995; Bullido and Valdivieso, 2000). Furthermore, polymorphisms in the proximal promoter and the first intron of the APOE gene cluster (−1,019 to +407) affecting APOE expression were identified in the late 90s and early 2000s (Mui et al., 1996; Lambert et al., 1997, 1998a,b, 2000; Bullido and Valdivieso, 2000; Lumsden et al., 2020). Importantly, these polymorphic sites have been associated with a differential risk of AD (Artiga et al., 1998; Sims et al., 2017). However, the association of those polymorphic sites, sequence variability in the proximal promoter and ME1 and ME2, and the level of APOE protein in AD are not clearly understood. The data on the expression level of APOE RNA and the correlation with APOE protein level differ across the studies, and the question why remains unanswered. Importantly, in human postmortem brain, a CGI island that overlaps with exon four and downstream is highly methylated, and the methylation is altered in AD frontal lobe (Lee et al., 2020). However, the methylation level of APOE CGI correlates to the expression level of four known APOE transcripts, of which only one is full-length. Surprisingly, circRNAs, miRNAs, and truncated APOE transcripts constitute a significant part of the total APOE mRNA with higher expression in the AD frontal lobe. While the precise clinical significance of these variations in the amounts of RNA and methylation level of CGI in the APOE 3′-exon are not fully understood, the results of an increasing number of studies point to the regulatory role of epigenomic signatures and changes in epigenome associated with risk or clinical presentation of a variety of neurological disorders (Lee et al., 2020).

For the last 25 years or so, the availability of a mouse model with a targeted replacement of mouse Apoe gene (Sullivan et al., 1997) provided an opportunity to address a myriad of questions ranging from the role of APOE isoforms in AD pathogenesis to how successful variety of therapeutic strategies have been so far. The model has been used in multiple studies aiming to reveal the role of APOE isoforms in the response to traumatic brain injury, as well (Crawford et al., 2002, 2009; Castranio et al., 2017, 2018). Considering the differences between mouse and human APOE gene clusters, how complex the transcriptional control of human APOE is, the structure of the targeting construct (s), and the strategy to replace mouse Apoe in the TR APOE models (Sullivan et al., 1997, 2009) become important factors. An NIH/NIA initiative recently established the MODEL-AD (Model Organism Development and Evaluation for Late-onset AD) consortium to provide the scientific community with AD animal models of LOAD that better mimic human disease (www.model-ad.org). The consortium has created knock-in humanized coding and non-coding LOAD risk variants expressed at endogenous levels, including mice expressing all APOE isoforms (Oblak et al., 2020; Kotredes et al., 2021; Foley et al., 2022). The targeting constructs used to generate APOE TR at Jax (Foley et al., 2022) differ from those originally used at Duke (Sullivan et al., 1997). In the Duke model the 3′ homology arm of the targeting construct, manipulated according to the classic genetic engineering technology, is upstream of the mouse Apoe exon 4 and 3′ noncoding sequences. That means no human regulatory sequences downstream of the human APOE gene exist in the chimeric gene. The targeting construct used in the Jax APOE TR mouse model was generated using a technology called recombineering (Sharan et al., 2009), which allows engineering of large constructs (>100 kb) and recombination events without leaving any ‘footprints’ behind homologous recombination. There are two differences between the targeting constructs/chimeric genes used in the Duke and Jax models: (1) in the Duke model the distal part of human APOE 1st intron, followed by the entire downstream sequences and part of 3′ sequences downstream of exon 4, followed by a Neo cassette with a stop codon are inserted between Sac I recognition site downstream of mouse 1^st exon/intron junction, and Pvu I recognition site within mouse Apoe exon 4 (Sullivan et al., 1997, 2009). (2) in the Jax model, the chimeric gene retains part of human regulatory sequences upstream of the noncoding exon 1and the entire genomic sequence of APOE. There is no Neo cassette and downstream of human APOE 3′ noncoding sequences the chimeric gene retains a small part of mouse distal noncoding exon 4 (Foley et al., 2022; Figure 1). Neither of the chimeric genes has human regulatory sequences downstream of APOE. While this is hard to predict with certainty, most probably the expression levels and response to external stimuli/insults in TR mice will not differ, regardless of the model. The availability of Jax APOE TR mice to the AD research community and the opportunity to generate, examine and compare APOEε3/ε4 heterozygous mice to APOEε3/ε3 and APOEε4/ε4 mice is what makes the models hugely different. So far, the patent restrictions have not allowed the generation of heterozygous TR mice. We can speculate, however, that the expression of APOE isoforms, in the brains of Jax TR mouse models, does not recapitulate the transcriptional control and the regulation of the expression of APOE isoforms in the human brain. While this poses some questions about how good the models are at studying the risk of AD, we can assume that in a mouse model without overexpression of human APP, different protein–protein interactions, receptor-ligand interactions, and downstream intracellular and extracellular effects, replicate true physiological or pathophysiological conditions responsive to regulatory mechanisms at various stages of a neurodegenerative disorder, including AD. We believe these are the principal arguments and a justification to conduct in vivo experiments using animal models expressing different APOE isoforms where mouse Trem2, for example, is physiologically expressed, globally deleted by genetic engineering, or expressed as a mutant form shown to be associated with AD. We expect that animal models allowing the analysis of APOE in the context of gene–gene and gene–environment interactions will appear soon.

In APP transgenic mice expressing human APOE isoforms, APOE affects Aβ clearance and deposition in an isoform-dependent manner (Holtzman et al., 2000; Bales et al., 2009; Castellano et al., 2011; Kim et al., 2011), and APOE lipidation level is of significance (Fitz et al., 2012; Liao et al., 2018). Data from ABCA1 deficient mice (Hirsch-Reinshagen et al., 2004; Wahrle et al., 2004) and Tangier disease patients with non-functional ABCA1 demonstrated that in the absence of cholesterol efflux plasma APOA-I is virtually missing and APOE in plasma and brain is significantly reduced [reviewed in Oram and Vaughan (2000) and Koldamova et al. (2014)]. It has been shown independently by three groups that in APP transgenic mice, global deletion of Abca1 decreases APOE lipidation and significantly increases amyloid deposition (Hirsch-Reinshagen et al., 2005; Koldamova et al., 2005; Wahrle et al., 2005). Recently, Fitz et al., using preclinical AD mouse models, demonstrated that APOE3 lipoproteins, compared to APOE4, prompted faster microglial migration toward injected Aβ, facilitated Aβ uptake, and abrogated damaging effects of Aβ oligomers on cognition (Fitz et al., 2021). Applying in vivo two-photon imaging, they showed that the APOE3 lipoproteins caused microglia to move faster toward Aβ and surround it, thus decreasing the spread of Aβ (Fitz et al., 2021). In tau mouse models of AD, expression of APOE4 has been shown to exacerbate tau-mediated neurodegeneration compared to mice with APOE3 expression, while the selective removal of astrocytic APOE4 is protective (Wang et al., 2021).

APOE and TREM2 are part of a large group of GWAS-identified genes associated with AD risk with glial-specific expression (microglia and astrocytes) and are related to immune response (Villegas-Llerena et al., 2016). As noted above, at least in vitro TREM2 binds to APOE (Atagi et al., 2015; Bailey et al., 2015; Yeh et al., 2016; Lessard et al., 2018; Kober et al., 2020). In some studies, the R47H variant of TREM2 has markedly reduced binding between APOE and TREM2 (Atagi et al., 2015; Bailey et al., 2015; Yeh et al., 2016) and in others with no effect on their interaction (Lessard et al., 2018). In terms of the affinity of TREM2 to bind different APOE isoforms, two studies showed that APOE4 demonstrates a slightly higher affinity to bind TREM2 than the other two isoforms (Jendresen et al., 2017; Kober et al., 2020) but other – no significant difference between the isoforms (Atagi et al., 2015; Bailey et al., 2015; Yeh et al., 2017; Lessard et al., 2018). Regarding APOE lipidation, TREM2 was shown to bind lipidated and non-lipidated APOE. However, it is an open question if APOE lipidation affects TREM2 binding. Some studies found that their interaction is enhanced by lipidation (Yeh et al., 2016) and others to be slightly decreased by it (Kober et al., 2020). As noted above, bearing in mind the data from Tangier patients and Abca1 knockout mice, it is possible that in vivo APOE does not exist in lipid-free form [reviewed in Oram and Vaughan (2000) and Koldamova et al. (2014)]. Thus, whether APOE lipidation has a role in TREM2 binding is more of a theoretical than practical significance.

Only a few reports examine the interaction between TREM2 deficiency and APOE isoforms in vivo either in APP transgenic models (Fitz et al., 2020) or in AD patients (Nguyen et al., 2020). Fitz et al. demonstrated that in APP mice expressing human APOE3 or APOE4 (APP/E3 and APP/E4), the lack of Trem2 impaired microglia barrier in both isoforms but did not change steady-state plaque load (Fitz et al., 2020). However, APOE mRNA expression measured in plaque-associated microglia was significantly reduced by Trem2 deficiency only in APP/E4 and not in APP/E3 mice, suggesting that APOE4 microglia respond differently to the absence of Trem2 (Fitz et al., 2020). Since another DAM gene – Clec7a, was also significantly decreased in plaque microenvironment only in APP/E4 without Trem2, this might imply that DAM response to a challenge (in this case, amyloid deposition and Trem2 absence) is less effective on an APOE4 background, thus preventing microglia around plaques to form a complete barrier (Fitz et al., 2020, 2021). A recent study by Nguyen et al. used single-nucleus RNA sequencing of postmortem human brain expressing APOE and TREM2 variants and identified four distinct microglia clusters. One was the homeostatic microglia; the rest were “active” clusters designated based on different pathology and differentially expressed genes. Nguyen et al. established that one of the active clusters includes a subpopulation of CD163-positive microglia, which they named “amyloid-responsive microglia.” They determined that this cluster is enriched in APOE3/3 AD patients and is relatively depleted in cases with APOE4 carriers and TREM2 risk variants. Nguyen et al. proposed an amyloid-responsive microglia subpopulation primed to elicit an activated immune response and note the reduced response in APOE4 carriers. In a recent study, Fitz et al. showed that in mice injected with Aβ plus native APOE lipoproteins, there was a higher number of differentially expressed genes between WT vs. Trem2^ko if the mice were injected with APOE4 compared to APOE3, particularly genes associated with interferon signaling (Ifit2, Ifi27l2a, Ifi207, and Axl) and endocytosis (Cd14, Cxcl16, Fth1, and Ifitm3) (Fitz et al., 2021). Additionally, the lack of TREM2 decreases Aβ phagocytosis only by APOE4-treated microglia, thus suggesting that APOE4 lipoproteins compared to APOE3 are insufficient to resist TREM2 deficiency, particularly in the presence of Aβ (Fitz et al., 2021).

It is hard to admit that 30 years after the discovery of APOEε4 as the highest genetic risk of AD, the nature and the molecular and cellular mechanisms that materialize the risk are still poorly understood. Multiple and well-supported hypotheses have been proposed, pointing to various mechanisms explaining the risk conferred by APOE4. Most importantly, however, loss of function – decreased level of APOE4 and fast degradation of poorly lipidated Apolipoprotein E4, as well as gain of function – generation of neurotoxic fragments due to domain interaction and reduced stability of APOE4, seem to work in concert and gradually lead to dysbalanced Aβ clearance, facilitated tau fibrillation and higher order behavioral disturbances. It is possible that these two pathogenic pathways work in concert and should be addressed together. A better understanding of those seemingly distant pathways and the interaction of APOE4 with other signaling and regulatory molecules – ABCA1 and TREM2, for example, conferring an increased risk themselves, are very important and hopefully will point to reasonable and probably successful therapeutic strategies based on APOEε4. Until then, the ideas of replacing/eliminating APOE4, inhibiting its interactions at an ill-defined age, or ignoring the intervention time seem poorly substantiated. In this review, we cautiously, although briefly, emphasized the complexity of transcriptional regulation of the E/C-I/C-IV/C-II gene cluster and differences in regulatory, including epigenetic, mechanisms in humans and mice. These differences become even more important in APOE TR mice. An attempt to formulate a strong hypothesis on how the risk conferred by the inheritance of the APOEε4 allele is materialized, based on the studies that use more or less complex or extremely complex animal models, is doomed to failure. The overwhelming controversies and inconsistencies in conclusions from otherwise perfectly conducted studies indicate the lack of the appropriate mouse model. Perhaps a model that includes APOE/CI part of APOE cluster with critical regulatory sequences is one promising option. The scientific community needs such a model/models to move further from square one – the inheritance of the APOEε4 allele is the highest genetic risk of AD.

IL and RK contributed to the conception of the review. YL wrote the first draft. IL, RK, and NF co-edited the manuscript. All authors contributed to the writing, have read and agreed on the final manuscript.

This work was funded by the National Institute on Aging – National Institutes of Health, USA: R01AG06619, R01AG077636, R01AG075992, R01AG057565, and R01AG052978.

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.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.