APOE Genetics as a Major Determinant of Alzheimer’s Disease Pathobiology

Keynote Lecture: APOE4 as a toxic gain of function molecule – Overview of background and recent findings on APOE’s multiple effects and roles in AD/ADRD

David M. Holtzman, M.D., Washington University

In the context of AD, Aβ begins aggregating and triggers an inflammatory response approximately 20 years prior to symptom onset. In contrast, the onset and progression of tauopathy tends to correlate with symptoms of cognitive decline in AD. However, researchers do not understand whether and through what mechanism Aβ drives tauopathy. In addition, APOE genotype is a major genetic risk factor for late-onset AD (LOAD). Relative to 2 copies of the E3 allele (APOE3), oneE4 allele (APOE4) increases risk of AD (OR=3.68) and one E2 allele (APOE2) reduces risk of AD (OR=0.64) in a dose-dependent manner. The Holtzman lab has been seeking to understand (1) how Aβ might drive tauopathy, (2) the role of inflammation in mediating Aβ and tau driven aspects of AD pathogenesis, and (3) how APOE mediates these processes.

Numerous genetic studies established APOE as a gene critical for modulating AD risk. One of the main ways that the APOE protein affects AD risk is through its interaction with amyloid. Depending on the APOE background, individuals exhibit premature or delayed amyloid deposition and accumulation. In a study of people between the ages of 50 and 70, approximately 60 percent of those homozygous for APOE4 (i.e., APOE4/E4) were amyloid positive, followed by 30 percent of those with the APOE3/E4 genotype, 25 percent of those with APOE3/E3, and 5 percent of those with APOE2/E3. However, previous genetic studies were primarily conducted in White European ancestry populations. More recent studies have identified differential APOE-related risk for AD across different ancestries. Ongoing research seeks to understand genetic factors that increase or decrease APOE-related risk of AD in different ancestral backgrounds.

APOE and Amyloid Deposition
Studies in mouse models of AD further confirmed APOE’s role in amyloid deposition. In amyloid-depositing mouse models, loss of APOE reduced amyloid deposition, and although some Aβ did accumulate in the absence of APOE, virtually none of it was fibrillar nor truly amyloid. Therefore, any APOE variant is likely important for amyloid deposition. In addition, transgenic expression of human APOE4 in astrocytes in APPV717F transgenic, Apoe knockout mice resulted in higher amyloid deposition relative to comparable mice expressing human APOE3. These data were confirmed using knock-in mice expressing human APOE2, APOE3, or APOE4 under control of the endogenous mouse Apoe locus. Studies comparing these three variants determined that soluble Aβ clearance was slower in APOE4 knock-in mice and faster in APOE2 knock-in mice relative to APOE3 knock-in mice. APOE also appears to directly impact aggregation of Aβ into amyloid deposits.

APOE and Tau-Mediated Neurodegeneration
APOE is required for cortical and hippocampal atrophy associated with tau pathology in PS19 mice. Normally, neurodegeneration in this mouse model occurs in brain regions with tau deposits. APOE4 expression in PS19 mice results in worse neurodegeneration compared to those expressing APOE3 or APOE2. APOE4 expression, in the presence of tauopathy, triggers an increase in inflammatory microglial genes and downregulation of microglial homeostatic genes, suggesting that APOE4 impacts neurodegeneration via immune pathways. Loss of APOE blocks this effect and so does removing microglia from the brain using a small molecule CSF1 receptor inhibitor, suggesting that microglia drive APOE-dependent neurodegeneration in this tauopathy model. Furthermore, T cell depletion, which eliminates the T cell-microglia interaction, also protects against brain atrophy. Therefore, how APOE interacts with the innate and adaptive immune responses in the context of AD remains an outstanding mechanistic question. Overall, based on mouse studies, APOE4 appears to exert a toxic gain-of-function (GOF) for main phenotypes relevant to AD including amyloidosis, tauopathy, inflammation, and neurodegeneration.

Protective APOE Variants
Examination of rare protective APOE variants such as APOE3-Christchurch (APOE3Ch) can provide further insight into the role of APOE in tauopathy and tau-mediated neurodegeneration. A prior case report described an individual carrying a presenilin 1 (PSEN1) autosomal dominant AD mutation with delayed symptom onset. This individual was homozygous for APOE3Ch and exhibited significant amyloid burden but low tau burden, which was very atypical anyone carrying this PSEN1 mutation. The individual did not develop symptoms of cognitive decline until age 70 when she was diagnosed with mild cognitive impairment (MCI). This is ~ 25 years after the mean age of mild cognitive impairment (MCI) onset in individuals with this PSEN1 mutation. Together, these observations suggest that some factor was blocking the progression from amyloid deposition to tauopathy.
Using various mouse models with humanized APOE3Ch Knock In (KI) mice, the Holtzman laboratory was able to determine whether APOE3Ch protected against specific aspects of AD pathology. In an amyloid induced tauopathy model—injection of tau in the brains of amyloid precursor protein (APP)/PSEN1-21miceexpressing humanized APOE3Ch, exhibited much less amyloid-induced tau seeding and spreading compared to humanized APOE3. Microglia and a microglial lysosomal marker were increased around amyloid plaques in this model expressing APOE3Ch compared to APOE3. To further interrogate the potential interaction between microglia and APOE3Ch, mouse myeloid cells from APOE3Ch or APOE3 KI mice were isolated and cultured with tau.. Myeloid cells from APOE3Ch-expressing mice took up tau more readily, degraded tau faster, and released less tau to re-seed back into the media compared to APOE3 myeloid cells. This finding implicates that microglial function in vivo as a critical step between amyloid deposition and tauopathy.

The APOE3Ch variant does not bind heparin sulfate proteoglycan, a co-receptor for low-density lipoprotein receptor-related protein 1 (LRP1) as effectively as APOE3 or other APOE variants, which suggests altered APOE regulated tau phagocytosis as both APOE and tau are ligands for the HSPG/LRP1 receptor pathway. By applying additional hAPOE3Ch protein to hAPOE3Ch myeloid cells co-cultured with paired helical tau filament, tau uptake was slightly reduced, while adding hAPOE3 protein significantly reduced tau uptake by these myeloid cells even further. Together, these results suggest that aggregated forms of tau may be competing with APOE for the same receptor systems on cells, including microglia, which partially impairs tau uptake and degradation. However, since APOE3Ch does not bind the LRP1 complex as effectively, tau can outcompete it, resulting in more efficient tau uptake and degradation. Additional data indicate that the effect of APOE3Ch on tau pathology is cell autonomous in microglia. Interestingly, expression of the Ch variant on an APOE4 background (APOE4Ch) in the PS19 mouse model and other tauopathy models resulted in less tau-mediated neurodegeneration compared to APOE4.

In addition to APOE3Ch, other rare APOE variants, including the Jacksonville variant (V236E) on an APOE3 background and the R251G variant on an APOE4 background are associated with reduced risk of AD. However, the mechanisms of the protective effects conferred by these variants are not completely clear and may or may not be related to the effects seen for APOE3Ch. Further studies of protective APOE variants and their potential profound effects on AD pathology are critical to understanding mechanistic relationships between APOE and AD.

Cell Type-Specific APOE Roles
APOE can exert cell autonomous and non-cell autonomous effects, both of which could be leveraged therapeutically. Astrocyte-specific knockout of APOE3 or APOE4 in APP/PSEN1-21 APOE knock-in mice reduces amyloid levels by 70 to 80%. This suggests that much of the APOE-driven amyloid deposition originates from APOE produced from astrocytes. In contrast, microglia-specific knockout of APOE4 influences disease-associated microglia (DAMs) around amyloid plaques, but has a relatively small effect on amyloid deposition. Therefore, APOE’s likely cell-autonomous role in microglia is modulation of microglial inflammatory state. Furthermore, selective expression of APOE3 or APOE4 in microglia in the absence of APOE subtly impacts amyloid deposition and further confirms cell autonomous effects on lipid metabolism, inflammatory gene expression, and lysosomal pathways in microglia.

Therapeutic Approaches
Based on mouse genetic studies showing that loss of one copy of APOE3 or APOE4 reduced amyloid deposition, reducing APOE3 or APOE4 levels may be a viable therapeutic strategy to reduce amyloid deposition. To test this hypothesis, antisense oligonucleotides (ASOs) targeting human APOE were injected into the brains of humanized APOE4/E4 and APOE3/E3 mice either at birth (prior to amyloid deposition) or a few months after birth (just after amyloid deposition). These ASOs lowered levels of APOE mRNA in the brain by 50 percent. Early injections resulted in reduced amyloid deposition and decreased dystrophic neurites comparable to that seen in single genetic copy loss experiments. However, when ASOs were injected after amyloid deposition began, amyloid plaque numbers were not reduced, but there were fewer dystrophic neurites per plaque compared to control mice. It is possible that a larger reduction in APOE expression may reduce amyloid plaques after initiation of amyloid deposition in mice but this needs to be tested. Further ASO experiments showed that reducing APOE4 in an APOE4 knock-in mouse model of tauopathy, TE4, reduced neuronal atrophy and death. This suggests that lowering APOE4 may be able to reduce tau-mediated neurodegeneration if done before significantly tau-related damage is present.

The potential protective effects of APOE2 suggest that overexpression of APOE2 could reduce amyloid pathology in amyloid depositing mice. To determine whether expression of APOE2 was sufficient to reduce amyloid deposition, APOE2 was expressed in ependymal cells of APP/PSEN1-21 APOE4/E4 mice after amyloid deposition began, allowing its secretion into cerebral spinal fluid (CSF) and subsequent delivery to other parts of the brain. This APOE2 expression exhibited a decrease in amyloid deposition over time in a dose-dependent manner, but further study is required to determine whether APOE2 actually clears amyloid deposits versus decreasing amyloid accrual. Overall, while evidence suggests that lowering APOE4 in the brain would be an important therapeutic target for AD, lowering or raising expression levels of other APOE variants requires further research.

Given that lipid accumulation is associated with tau-mediated neuronal degradation and that APOE may have a role in lipid accumulation in glia, further mouse studies were used to determine whether reducing lipid accumulation could be neuroprotective. Treatment with a liver X receptor (LXR) agonist in TE4 mice reduced lipid accumulation in glia, tauopathy, inflammation, and neurodegeneration. Overexpression of ABCA1—a gene previously implicated in AD and lipid transport—or of low-density lipoprotein receptor (LDLR) yielded similar protective effects. Together, these results suggest that reduction of lipids in the brain, particularly in microglia could also serve as a viable therapeutic strategy for AD.

Relationship Between APOE and Cerebral Amyloid Angiopathy
The anti-amyloid antibodies that are FDA approved therapeutics, lecanemab and donanemab, are associated with amyloid-related imaging abnormalities (ARIA), which have pathological features that are linked to cerebral amyloid angiopathy (CAA). CAA is the accumulation of Aβ in cerebral blood vessels. Notably, APOE4 is a significant risk factor for CAA. Functional studies in mice have identified a potential link between APOE and CAA. In transgenic mice expressing APP with the Swedish mutation, APOE is required for CAA formation. APOE status may also influence the ratio of parenchymal amyloid deposition and CAA severity. Some transgenic APP mice expressing mouse Apoe normally develop both parenchymal amyloid deposits and CAA. In comparison, transgenic APP mice expressing human APOE4 at the endogenous mouse Apoe locus have fewer parenchymal amyloid deposits but more substantial CAA. Alternatively, transgenic APP mice expressing APOE3 exhibit very little CAA in relation to parenchymal amyloid plaques. Further research is necessary to understand how APOE mechanistically can influence CAA.

Recently, a specific anti-APOE antibody, HAE-4 was shown to reduce CAA and parenchymal amyloid plaques and improves cerebrovascular function in transgenic APP mice. The monoclonal antibody used in these mouse experiments binds to an APOE conformation present in amyloid plaques and CAA but not the normal physiologic conformation of lipidated APOE. Notably, peripheral administration of anti-APOE immunotherapy (HAE-4) decreased CAA and parenchymal plaques, while an anti-amyloid immunotherapy (chimeric version of aducanumab) only reduces parenchymal plaques and can cause in increase in CAA-related hemosiderin, an ARIA-H like phenotype). Additional studies are needed to understand why this anti-APOE immunotherapy does not cause ARIA-H.

Discussion

Knockdown of both APOE and apolipoprotein J (APOJ) results in increased parenchymal amyloid deposition. How should we be thinking about amyloid deposition in the context of non-APOE apolipoproteins?

APOJ may have similar effects as APOE on Aβ metabolism, clearance, and aggregation. If more than one apolipoprotein is involved in this process, researchers need to understand the complexity of what regulates clearance. Knocking down one apolipoprotein may be beneficial therapeutically, but the manipulation of multiple apolipoproteins simultaneously presents additional complexity.

Is there evidence that APOE4 primes astrocytes and microglia to be chronically reactive at the earliest stages of AD?

In mouse models that express only APOE2, APOE3, or APOE4, astrocytes and microglia do not appear phenotypically different from one another early in life. It is possible that APOE variants start to result in differences in glia reactivity as AD-like pathology emerges. However, whether APOE impacts astrocytes and microglia early in life in humans or only after AD pathology emerges remains unclear. Notably, environmental exposures may also impact astrocyte and microglia priming in an APOE4-dependent manner. APOE4 allele frequency is highest in equatorial regions of Africa and South America, where people are exposed to a multitude of parasites and some bacteria; perhaps APOE4 is selected for in these regions because of some protective benefit relevant to these unique immune exposures.

The APOE4 Story: From Discovery to Diversity

Margaret A. Pericak-Vance, Ph.D., University of Miami

A 1984 Leadership in AD Award from NIA supported research on genetic risk factors for Late Onset Alzheimer’s Disease (LOAD). Unlike autosomal dominant AD pedigrees, LOAD pedigrees did not indicate a simple genetic linkage. Expansion of these pedigree analyses to nonparametric sibling pair analysis did not provide sufficient study power to identify genetic linkages due to low sample size. However, the affected-pedigree-member method of linkage analysis identified linkage to chromosomes 19 and 21. Further analysis failed to identify individual candidate genes, but after APOE—a gene on chromosome 19—was implicated in cardiovascular disease, subsequent analysis identified a genetic linkage between APOE and AD. Despite identifying this genetic linkage, LOAD was not thought to have a genetic component until 1993, when a publication reported an APOE4 allele risk effect on LOAD in a gene dose-dependent manner. In 1994, APOE2 was found to have a protective effect against LOAD.

Additional research found that APOE4/E4 homozygosity was sufficient to cause AD by age 80 with nearly full penetrance as well as predictable symptom onset, biomarker changes, and clinical changes. Early studies of APOE-related AD risk were conducted in people of European ancestry and thus failed to identify differential effects of APOE on AD risk across ancestral populations. However, in the 1990s, researchers conducted similar analyses on non-Hispanic White, Japanese, Hispanic, African American, Nigerian, and Chinese ancestries. Studies comparing differential APOE-related risk initially contained small sample sizes and large confidence intervals, but more recent, larger studies have tightened those confidence intervals and confirmed these ancestral differences, particularly a higher risk for AD associated with the APOE3/E4 genotype in Japanese and Korean ancestries and a lower risk in East and West African populations. However, additional, large-scale studies are essential to understanding the variation in APOE-related risk within these ancestral populations. For example, the DAWN study will analyze APOE-related risk in populations across 10 African countries.

Discussion

Have studies of ancestral effects on APOE-related AD risk examined Asian ancestries beyond Japanese and Koreans?

Other studies have also addressed the Chinese ancestral effects on APOE-related AD risk. More studies are needed in other Asian and Hispanic populations due to their inherent ancestral heterogeneity to fully dissect ancestral effects on APOE-related AD risk.

Local Versus global ancestry in APOE: African and African American, Hispanic, and Asian Populations

Farid Rajabli, Ph.D., University of Miami

The association of the APOE3/E4 genotype and AD risk varies across different ancestral populations, with the highest associated risk in Japanese and Korean ancestries and lowest associated risk in West and East African ancestral populations. To better understand how ancestral background affects APOE-related AD risk, we employ both global and local ancestral approaches. Across the world, populations that were once historically isolated have since come together and integrated into new populations, and this is known as admixture. While different racial and ethnic groups as well as countries have varied proportions of African, European and Amerindian ancestries (i.e., global ancestry), the distribution of these ancestries is not evenly distributed across chromosomes. Therefore, when assessing AD risk in different populations, we can stratify people within a population based on chromosomal distribution of different ancestries and ancestral background at a given genetic region of interest (i.e., local ancestry).

Based on local ancestry-based fine mapping of risk heterogeneity, ancestry-specific genetic factors near APOE contribute to the lower risk effect of APOE in African Americans and Puerto Ricans with local African ancestry compared to European ancestry. Among Caribbean Hispanics, individuals with African ancestry at APOE4 had a 39 percent lower odds of AD than individuals with European-derived APOE4. Furthermore, among a Brazilian admixed population, lower neuropathological burden was associated with African ancestry at the APOE4 locus compared to European ancestry.

Current studies on the ancestral effects on APOE-related AD risk focus on a wide range of countries and ancestries, including Peru with a significant amount of Amerindian ancestry, West Africa with mostly African but also Middle Eastern and European ancestry, and East Africa with African ancestry with significant admixture of Middle Eastern ancestry and some admixture of South Asian and European. The odds ratios (ORs) for APOE genotype-related AD risk are outlined in the table below (unpublished). Notably, while APOE4/E4 confers higher risk compared to APOE3/E4 across these countries, due to the current small sample sizes, the confidence intervals for these ORs are fairly large.

Local ancestry stratification of participants in Peru indicated that local Amerindian background contributes to a higher OR than the European background. The local ancestral analyses of participants in West Africa found that those with nearly or all African ancestry had no significant association between APOE4 and AD risk. Some of these participants have admixture of European and Middle Eastern ancestries, which could account for the association between APOE4 and AD in West Africa. Notably, participants from East Africa had significantly different admixture depending on the country. Those in Ethiopia and Kenya have significant and uniform amounts of Middle Eastern ancestry. Those in Mozambique have varied admixture with some South Asian, European, and Middle Eastern ancestries. Functional follow-up studies using local ancestry have shown that different APOE4 expression is associated with the difference in AD risk across different ancestral backgrounds.

Discussion

Why does the East African population have such a higher APOE-related AD risk compared to the East African ancestry?

Data from Africa are very recent and due to small sample size, the confidence intervals (CI) for ORs are very wide. In our study only 12 individuals from East Africa had the APOE4/E4 genotype, so more data are needed to fully understand the true effect of that genotype. Notably though, a pattern is emerging that APOE4/E4 confers an exceedingly high risk for AD across different ancestral populations, while APOE4 heterozygotes have different risk levels across populations. For example, in Yoruba, which has almost 100 percent African ancestry, the APOE3/E4 genotype has little to no effect on AD risk, while in Ghana and other countries with some Middle Eastern and European admixture the genotype has a significant effect on AD risk, suggesting that African ancestry may have a protective effect against APOE-related AD risk.

When comparing APOE-related AD risk across ancestries, did you consider lifespan?

Lifespan differences are worth considering as a factor that may impact APOE-related AD risk observed across different populations.

In Europe, there is a gradient of AD risk from North to South. How much of this differential risk is due to genetics and ancestry versus environment?

There are differences in the effect size of APOE-related AD risk in Northern versus Southern European populations. Those risks do seem to be inherited within those populations, suggesting a correlation with genetic ancestry. However, environment should always be considered as an additional factor for AD risk when looking across different populations.

Differential APOE effects on gene expression in different human brain cell types

Anthony J. Griswold, Ph.D., University of Miami

APOE expression levels vary depending on cell type, with APOE expressed highest in the liver and adrenal glands, likely due to its roles in lipid metabolism, as well as in brain tissue. Single-cell RNA sequencing (scRNA-seq) determined that APOE is highly expressed in microglia and astrocytes from the cortex, moderately expressed in oligodendrocytes and endothelial cells, and minimally expressed across other brain cell types. Furthermore, in a set of brain samples from various APOE backgrounds, APOE was more highly expressed in astrocytes and microglia in pathologically diagnosed AD compared to those without neuropathology. Additional scRNA-seq was used to determine (1) whether cell type-specific APOE expression varied based on APOE genotype, (2) whether ancestral background impacted cell type-specific APOE expression levels, and (3) how APOE genotype and local ancestry may regulate APOE expression.

Overall, APOE4/E4 brains express 1.8-fold more APOE in the brain compared to APOE3/E3 (unpublished). In addition, individuals with European local ancestry at the APOE locus in APOE4/E4 genotype expressed nearly two-fold more APOE in microglia and astrocytes compared to those with African local ancestry at the APOE locus in APOE4/E4 genotype. Based on assay for transposase-accessible chromatin with sequencing (ATAC-seq) data, the increased expression in European local ancestry was partially due to increased chromatin accessibility at the APOE promoter.

To further dissect differences in regulatory elements of APOE in European versus African local ancestry, genome fragments near APOE across haplotypes with different frequencies in European and African ancestries were cloned and fused to a reporter to determine whether these fragments can drive expression using a massively parallel reporter assay platform. Individual luciferase reporter assays identified regions in two introns in a nearby gene, translocase of outer mitochondrial membrane 40 (TOMM40), confirming that some genome fragments near APOE can drive its expression. Based on promoter capture Hi-C analysis of chromatin architecture, these regions loop back to the APOE promoter, suggesting direct regulation of APOE expression. Removal of one of these regions in microglia derived from an induced pluripotent stem cell (iPSC) line of African ancestry resulted in increased APOE expression, suggesting that this region is possibly a transcriptional repressor (unpublished). Similar methods are being used in both microglia and astrocytes to identify additional regulatory elements.

Our laboratory will continue to analyze chromatin conformation using Hi-C and high-throughput regulatory element modulation using Perturb-seq to further dissect ancestry-specific regulatory mechanisms of APOE expression. While current work has focused on APOE3 and APOE4, future studies should also consider other APOE genotypes as well as other ancestries, including Amerindian and Asian. Ultimately, this work will provide further insight into how to safely reduce expression of APOE4 to prevent and treat AD.

Discussion

Does the level of AD pathology impact baseline regulation of APOE expression in different cell types?

Brain tissue used in the scRNA-seq experiments were from AD donors, while a lot of other scRNA-seq datasets are comprised of mixes of AD cases and controls. In the experimental design phase, when comparing different APOE genotypes or ancestries, pathological AD staging is used to ensure similar disease staging between groups. Future work will include scRNA-seq in tissues prior to onset of AD symptoms as well as scRNA-seq datasets across ancestries. Notably, the majority of autopsy materials are from non-Hispanic White donors, so additional material is needed from non-European ancestries. Another set of experiments we are working on will leverage iPSC-derived oligodendrocytes, microglia, astrocytes, and neurons from different ancestral backgrounds to compare expression differences.

Is the enhancer element you identified microglia- and astrocyte-specific or is it also important in the periphery?

Whether this enhancer is important in the periphery remains unknown. Lowering APOE levels throughout the body could exert detrimental effects in the periphery. However, in principle, single nucleotide polymorphism (SNPs) found in brain tissue should exist in all cell types, but whether these SNPs also have consequences for APOE expression in other cell types remains unclear.

When looking at APOE expression in microglia and astrocytes across different ancestries, are APOE levels different across different subclasses of microglia and astrocytes?

In European backgrounds, compared to African backgrounds, a cluster of astrocytes in the scRNA-seq data appear activated and exhibit the highest expression of APOE among astrocytes. However, given the low number of cells in the dataset, comparisons of cellular subtypes were currently not possible. As sample sizes increase, potential differences in cell subtypes should be interrogated.

Have you considered whether other genetic modifiers contribute to the ancestry effect on APOE-related AD risk?

When looking globally in the genome, distant genes and regions do not seem to contribute much to APOE-related AD risk. The local ancestry near the APOE locus appears much more important for modulating this risk.

Does APOE mRNA expression correlate with protein levels of APOE isoforms in the brain?

APOE protein is made in both astrocytes and microglia, but whether protein expression differences exist between APOE3 and APOE4 in these two cell types remains unknown. However, iPSC models from different APOE backgrounds can be leveraged to measure cell type-specific protein expression differences.

Are the differences in APOE isoform expression present at basal levels, or are these differences related to some environmental trigger related to AD?

This question of cause and effect requires further investigation.

How much variation is there in the APOE4 haplotype across ancestries?

Based on regulatory region and expression differences, haplotypes differ widely across ancestries. However, long read sequencing in different ancestries will provide more granular information on these haplotype differences.

Are the differential effects on APOE-related AD risk explainable by individual SNPs or a more collective haplotype?

Hi-C data indicate a significant amount of interaction with the APOE promoter, suggesting a lot of transcriptional control, which makes sense, given that APOE is important for stress responses and other cellular mechanisms. APOE expression is likely regulated via a larger haplotype effect with more than just one enhancer. In addition, given that APOE is most highly expressed right after birth, further research is essential to understand the regulation of APOE expression over time and how this affects AD risk.

Dissecting APOE biology by CRISPR-based functional genomics

Martin Kampmann, Ph.D., University of California, San Francisco

Dr. Martin Kampmann uses unbiased clustered regularly interspaced short palindromic repeats (CRISPR)-based functional genomic screens in iPSC-derived brain cell types to identify genetic modifiers and uncover mechanisms of neurodegenerative diseases. These screens leverage a dead Cas9 enzyme fused with transcriptional activator or repressor domains to increase or decrease target expression, respectively. Paired with a three-dimensional cell culture system (iAssembloid) containing separately iPSC-derived neurons, astrocytes, and microglia, these CRISPR screening methods can perturb genes in a cell type-specific manner. Notably, these iAssembloids likely reflect more mature neuronal phenotypes compared to other two-dimensional cell culture systems, with greater maturity based on expression profiles and higher electrical activity.

To identify modifiers of neuronal survival in APOE3 versus APOE4 glial environments, Dr. Kampmann conducted a CRISPR screen using a guide RNA library against 2,300 genes, representative of the druggable genome, in iPSC-derived APOE3/E3 neurons. These neurons were co-cultured with APOE3/E3 or APOE4/E4 astrocytes to identify genetic perturbations in neurons that impacted neuronal survival in these APOE astrocyte backgrounds. This screen was conducted in two different genetic backgrounds—KOLF2.1 neurons plus KOLF2.1 APOE3/E3 or APOE4/E4 astrocytes and WTC11 neurons plus isogenic APOE3/E3 or APOE4/E4 astrocytes—to identify modifiers in common.

While this CRISPR modifier screen detected multiple genes that impacted neuronal survival regardless of APOE glial environment, calcium voltage-gated channel auxiliary subunit beta 4 (CACNB4) had differential effects on neuronal survival in APOE4 versus APOE3 glial environments. CACNB4 is a subunit of a synaptic voltage-gated calcium channel, and its loss of function is linked to epilepsy, suggesting a potential link to neuronal hyperexcitability. Subsequent in vitro experiments validated this CACNB4 finding. Knockdown of CACNB4 in APOE3 neurons in an APOE3 astrocyte background was not toxic to neurons. However, knockdown of CACNB4 in APOE3 neurons in an APOE4 astrocyte background was somewhat toxic to neurons.
Dr. Kampman further investigated potential mechanisms for how CACNB4 could have a differential effect on neuronal survival in APOE4 and APOE3 astrocyte backgrounds. First, neurons cultured in an APOE4/E4 glial background are more active and have higher reactive oxygen species (ROS) levels compared to those cultured in an APOE3/E3 glial environment. Importantly, ROS and neuronal hyperactivity can lead to lipid peroxidation and ferroptosis. Blocking neuronal hyperactivity lowers ROS levels, while blocking ferroptosis improves neuronal survival. Notably, increasing the ratio of astrocytes to neurons in culture can counteract neuronal hyperactivity and ROS. Therefore, the identified CACNB4 mutation may impact neuronal survivability via neuronal hyperexcitability.

Future work on understanding neuronal survivability in APOE4/E4 versus APOE3/E3 backgrounds will include mouse in vivo CRISPR screens using adenovirus-associated vectors (AAVs) to target neurons and astrocytes. This method enables perturbation of thousands of genes within a single mouse brain.

Discussion

Do you think the phenotypic differences seen between APOE4 and APOE3 could be driven in part by loss of prosaposin (PSEP) function, and therefore lysosome dysfunction?

PSEP was not targeted using the initial 2,300 guide RNA library, but future genome-wide CRISPR studies will help determine whether these APOE4 and APOE3 differences are due to lysosomal function. Intriguingly, PSEP has been identified in many other genome-wide modifier screens, including those looking for modifiers of tau aggregation. PSEP perturbation seems to worsen tau aggregation.

Both cell lines used are derived from males. Given gender effects on AD, will you validate these studies in female-derived cell lines?

Repeating these experiments in multiple male- and female-derived iPSC lines is essential to assess the validity and generalizability of these results. In vivo screening in mice would also allow for examination of sex differences. Drs. Li Huei Tsai and Vance offered to share isogenic APOE3 and APOE4 iPSC lines derived from females and different ancestries, respectively.

For the phenotypes described, is the physical presence of astrocytes necessary, or is the conditioned media from astrocytes sufficient?

An experiment testing the sufficiency of conditioned media is essential. Astrocytes may serve a buffering function for concentration of glutamate and possibly other factors.

Have you also conducted these screens to investigate potential roles of astrocyte APOE background on neuronal survival?

APOE3 and APOE4 astrocytes in this system do exhibit differences on neuronal survival, and future work aims to identify potential mechanisms. In addition, the APOE background of neurons may also impact neuronal survival and astrocyte phenotype.

Do you see an ELOVL1-related phenotype? Are the neurotoxic effects observed in this study specific to a certain lipid species carried by APOE?

Long-chain saturated fatty acids in media lead to some of these observed effects. Currently, the laboratory is trying to recapitulate Dr. Shane Liddelow’s (New York University) findings in this iAssembloid model. Conditioned media from reactive astrocytes does exert neuronal toxicity, and removal of lipids from the media reduces this toxicity. However, the laboratory has not investigated lipid species differences between APOE3 and APOE4 astrocytes. Notably, because Dr. Liddelow did not report induction of neuronal hyperexcitability nor heightened ROS, his results may be indicative of a different mechanism parallel to neuronal hyperexcitability. In the case of the Kampman laboratory’s findings, the observed neuronal toxicity may result from loss of the protective function of astrocytes rather than from astrocytes gaining a toxic function, because removal of astrocytes from the iAssembloid system results in worsened neuronal hyperexcitability and heightened ROS levels.

South Asians in India (LASI-DAD): Impact of APOE4 with Social Determinants of Health (SDOH)

Sharon L. R. Kardia, Ph.D., University of Michigan

The population of India is genetically and socially diverse, but prior genetic studies in the country relied on geographically-restricted and nonrepresentative samples. Therefore, the Harmonized Diagnostic Assessment of Dementia for the Longitudinal Aging Study in India (LASI-DAD) seeks to study late-life cognition and dementia in a representative subset of Longitudinal Aging Study in India (LASI) participants. LASI is a nationally representative survey of health, economic, and social wellbeing of an over 73,000-person sample of the Indian population age 45 and older. LASI-DAD enrolled 4,096 of these participants who were 60 years and older—2,700 of whom have whole-genome sequencing (WGS) data—oversampling those at high risk of cognitive impairment based on cognitive test performance and proxy interview results from the main LASI study. LASI-DAD pairs cognitive assessments with rich demographic, social, dietary, environment, and biomarker measurements to better understand cognition in late-life.

Using LASI-DAD data to understand the effects of APOE4 on AD risk in India requires a deep understanding of the region’s heterogenous genetic structure and its complex social environment. The Indian population experiences differential exposures to air pollution and exhibit varied health behaviors, as well as different languages, religions, and cultures. Importantly, assortative mating based on caste and tribe has resulted in homozygosity and founder effects. Due to these complexities, the effects of APOE4 on cognitive function are unclear in the Indian population, as are potential modifications of risk by sociodemographic factors such as location and education.

LASI-DAD data depicted higher than expected micro-genetic-heterogeneity for APOE status as well as unexpected APOE4-environment relationships, further complicating studies of APOE and AD with additional confounders. APOE4 carriers were associated with lower education, illiteracy, rural residence, be in a scheduled tribe, have lower income, and have lower cognitive scores. To account for these potential confounders inherent to the Indian social structure, the LASI-DAD dataset needs to grow significantly, including additional WGS data, ideally for the full 70,000 LASI cohort.

LASI-DAD data also uncovered a large amount of variation in cognitive measures and that sociodemographic variables could explain a significant amount of this variation. Sociodemographic variables explaining this variation include sex, age, state, education, literacy, and rural versus urban residence. In addition, genetic principal components explain additional variation even after accounting for these sociodemographic variables. Combined, these factors can explain between 40 to 60 percent of variation in cognitive measures. Furthermore, APOE4 status was associated with most cognitive measures, explaining between 0.1 to 0.5% of the variability. Each copy of APOE4 was associated with approximately one point reduction in Hindi Mental State Examination score and about 0.1 standard deviation reduction in other cognitive scores. These APOE4 effects are more pronounced with age, and females exhibit stronger APOE4 effects compared to males. These effects were also slightly stronger in those without a formal education; however, sex may confound this finding given that most participants without formal education were females.

Discussion

Are you stratifying your analyses based on dementia status? Which biomarkers will you use to determine AD pathology?

Currently, LASI-DAD researchers have been working with clinicians to assess the whole cohort for AD symptoms. They have taken a combination of approaches considering the complexity of AD/ADRD diagnosis and dealing with challenges in more remote areas. A subsample of LASI-DAD participants (N=600) received MRI scans and all consented LASI_DAD participants were assessed for some relevant blood-based biomarkers ((e.g., beta amyloid 40, 42, t-tau, p-tau 181, GFAP, NFL). Based on the Wave 1 data, approximately 7.4% percent of the LASI-DAD sample has dementia.

Could exploring APOE4 carrier status by state reveal microheterogeneity caused by ethnic customs in marriage and India’s caste system?

LASI-DAD currently has too small of a sample size and is too underpowered to answer this question. If the LASI-DAD cohort were expanded to include all 70,000 LASI participants, this sample size could be sufficient to dissect genetic contributions from customs related to within-group marriages, certain diets, and overall lifestyles.

What was the variation in the OR for APOE4-related risk of AD?

The effect of APOE status on AD risk varies widely by state. While each state may only have 100-200 participants, these cohort sizes per state are sufficient to provide some insight into APOE-related risk trends.

Have you looked at lipid levels in this population, and how much do lipid levels vary across different subpopulations?

This population has wide variation in lipid levels across subpopulations, especially low density lipoprotein cholesterol, but this variation is not as large as the variation seen in cognitive measures. State explains 3.9%, 4.1%, 6.1%, and 5.2% variance for total cholesterol, hdl, ldl and triglycerides. The corresponding variance for cognitive measures explained by state ranges from 5.3% to 18.2%

In your observed association between APOE and education, did you see an association between APOE4 frequency and education or that education modifies the effect of APOE4 on cognition?

APOE-related risk of lower cognitive function is strongest in those with no formal education.

Will you be analyzing any biomarker data for LASI-DAD?

Biomarker data analyses are currently in progress. LASI-DAD team completed the biomarker assay for both Wave 1 and 2 samples, including the full 2,700 LASI-DAD cohort.

A Genetic Modifier of ε4/AD Association and APOE Expression

Lindsay Farrer, Ph.D., Boston University

The effect of the APOE4 allele on AD risk is highly variable across major ancestral populations, with APOE4 conferring the highest risk for AD in East Asian populations, followed by European and then African American populations. The basis for these population differences is unknown, but several hypotheses have been proposed including the presence of modifiers within or extant to the APOE locus or moderating influences of dietary and environmental factors. Therefore, we analyzed SNP data to identify potential modifiers within the APOE gene and nearby regulatory regions. We hypothesized that the ethnic differences in the effect size of the ε4/ε4 genotype on AD risk may be due to the moderating effect of variants in the APOE region that have different frequencies across ethnic groups. An initial screen of the 5.7 kB region within and surrounding APOE identified 57 SNPs, which required further prioritization for additional research. Using European, African American, and East Asian data—including data on South Koreans from the Gwangju Alzheimer’s and Related Dementias Study (GARD). We narrowed down the number of priority SNPs to three, excluding the APOE4 variant. Candidate SNPs needed to (1) be present at an allele frequency higher than 1 percent in East Asian, European, and African ancestries, (2) be increasing or decreasing in frequency in the order of East Asian, European, and African, and (3) have at least a five percent difference in allele frequency between ancestral groups.

One SNP—rs405509—present at the -219 position in the APOE promoter, exhibited a modifying effect on the APOE-related risk of AD. In East Asian and European ancestries, the T/T genotype at rs405509 conferred a higher risk of AD among APOE3/E4 and APOE4/E4 individuals, followed by the T/G and G/G genotypes in those APOE backgrounds. This differential risk was also discernable based on age of onset of AD in East Asian and European ancestral populations. Using CRISPR in vitro to create APOE promoter constructs with T or G alleles fused to a luciferase reporter resulted in reduced luciferase expression under control of the T allele promoter compared to the G allele promoter in vitro. This effect on expression was also confirmed in both human cortex and cerebellum samples, although the reduced expression due to the T allele was evident on both APOE E4/E4 and E3/E3 genotype backgrounds. A multi-tissue expression quantitative trait locus comparison using Genotype-Tissue Expression data also found that while the G allele had increased expression in human cortex, cerebellum, and whole blood, it had an opposite expression effect in other tissues, further complicating potential modifying mechanisms for this SNP. In human dorsolateral prefrontal cortex tissue from ROSMAP brain donors, those with the T promoter allele trended toward increased expression compared to the G promoter allele, but only in AD cases, and mostly driven by donors who did not have the APOE4 allele.

Overall, Dr. Farrer posited how to reconcile the seemingly incongruent findings of the rs405509 T allele increasing AD risk among APOE4/E4 individuals while also lowering APOE expression. Because the rs405509 T allele reduces APOE expression regardless of APOE genotype, its modifying effect on the APOE4-AD risk relationship could be through some other mechanism besides APOE expression levels. This hypothesis is supported by the following observations: (1) the T allele was associated with increased APOE expression in AD but not control brains and (2) the effect of the T allele on AD risk appears independent from APOE4 status.

Discussion

Why do APOE4/E4 homozygotes have larger confidence intervals compared to APOE4 heterozygotes?

The difference in confidence intervals is mostly driven by small sample size.

What effect does the T allele have on APOE in different cell types? What effect does the T allele have on APOE expression in AD cases?

The effect of the T allele on expression varies both in magnitude and direction (i.e., increased or decreased) of APOE across different cell types. The effect of the T allele on cell type-specific APOE expression in AD cases has not yet been studied.

Dr. Vivian Chung’s laboratory at University of Michigan has been investigating this SNP and actually thinks it is in an enhancer region in a separate cap RNA, not the promoter.

Dr. Vivian Chung noted that the transcription pattern of the region containing this SNP could explain why in some contexts it may not act like a typical enhancer. The enhancer function is very cell type-specific in terms of its function in response to stress. It forms an R loop, and the RNA in the loop is modified, regulates the enhancer itself and by extension, APOE expression.

Are APOE protein levels in plasma, CSF, or brain homogenate related to this modifier SNP?

Data from the Framingham Heart Study could be used to answer this question.

What is the linkage disequilibrium between this SNP and APOE SNPs?

The linkage disequilibrium is not high between this SNP and APOE SNPs.

Is the reduction of APOE expression having an effect separate from an APOE4 GOF toxicity?

The answer to this question remains unclear, especially given the limited phenotype data across the dataset.

PSG haplotype is protective for APOE4

Jeffery M. Vance, M.D., Ph.D., University of Miami

APOE risk for AD differs between populations, and factors contributing to lower risk effect of APOE are likely due to ancestry-specific genetic factors near APOE rather than non-genetic ethnic, cultural, and environmental factors. One such genetic factor is located in a locus at 19q13.31 that significantly reduces the APOE4-related AD risk in African ancestry. The rs10423769 A allele reduces risk for AD in African ancestry with the APOE4/E4 background by about 75 percent. This A allele is present at a 12 percent frequency in African and only 0.3 percent frequency in European ancestries. The rs10423769 A allele lies within a region of segmental duplication and is located 2 Mb upstream of APOE. These segmental duplications encode cluster of pregnancy specific beta-1 glycoproteins (PSG) genes and long noncoding RNAs primarily expressed in the placenta. According to AD Sequencing Project (ADSP) database data and nanopore long-read sequencing and short-read sequencing data, this A allele belonged to a unique haplotype not duplicated in another region of the genome. The rs10423769 A allele haplotype is found in multiple African populations at a frequency of at least nine percent.

To better understand how the rs10423769 A allele haplotype protects against AD, long read sequencing was used to identify structural variants (SVs). While no SVs were identified in this immediate haplotype, an insertion in a tandem repeat region of chromosome 19 was identified as having high linkage disequilibrium with the A allele haplotype and was rarely seen in those with the G allele. This region contains the binding motif for myocyte enhancer factor 2 (MEF2) transcription factor and is located 32 kB from the A allele. Local assembly of this tandem repeat region revealed that the A allele is also associated with expanded variable number of tandem repeat (VNTR) alleles. VNTR length is much greater in the A allele haplotype compared to the G allele. While the association between the A allele haplotype and the expanded VNTR is strong, how the VNTR affects APOE-related risk remains unclear. While allele-specific differential methylation is evident between these two alleles, the locus is far from the APOE promoter, although prior studies have shown that methylation can impact expression at high distances. Hi-C compartment analysis data suggested that those with both the A allele and African local ancestry at APOE had reduced expression of APOE. These results were further confirmed using bulk RNA sequencing (RNA-seq) of cerebellum and frontal pole tissues from human brains. An allele carrier brain had the lowest APOE expression in both tissues compared to G allele carriers, and scRNA-seq is in progress to identify cell type-specific differences.

Next steps to better understand these haplotypes will further characterize segmental duplication differences between A allele and G allele haplotypes as well as cell type-specific APOE expression differences. Due to high similarity across these segmental duplications, PacBio HiFi sequencing at a depth between 40 and 60 is necessary to generate assemblies of this region. Contigs will also be cloned between the PSG region and the APOE locus. iPSC studies of lines derived from donors of African local ancestry carrying the A or G allele will enable further functional studies to understand potential protective mechanisms.

Discussion

Since the A allele is associated with lower expression but a longer VNTR, could there potentially be a repressor element in this region?

The region could contain a repressor element or could be an enhancer present in European ancestry and absent in African ancestry. Generally, Hi-C and ATAC-seq data suggest that the European chromosome has significantly more open chromatin than the African chromosome, so there could be a chromatin effect.

Fibronectin 1 and APOE-ε4

Richard Mayeux, M.D., Columbia University

Some individuals with the APOEε4/4 genotype do not develop Alzheimer’s disease as measured by changes in cognitive performance. This phenomenon was first observed in the Dominican population and later in the Washington Heights population. These individuals also had normal levels of cerebrospinal fluid (CSF) biomarkers associated with AD, suggesting potential protection against the effects of APOEε4. This protective effect may be due to genetic variants that counteract the impact of APOEε4. To investigate this further, Dr. Richard Mayeux coordinated a family-based study across the United States, analyzing whole-genome sequencing (WGS) data from 3,578 individuals. The study included: (1) APOEε4 heterozygotes aged 80 or older with no signs of dementia, (2) APOEε4 homozygotes aged 75 or older with no signs of dementia, (3) APOEε4 carriers with AD, and (4) non-APOEε4 carriers with or without AD. The hypothesis was that protective genetic variants might exist in APOEε4 carriers without AD but would be absent in APOEε4 carriers with AD. An initial analysis of these genomes identified 80 million genetic variants, which were then filtered to exclude variants found in APOEε4 heterozygous or homozygous individuals with AD. This filtering process resulted in a total of 510 candidate protective variants. Dr. Badri Vardarajan’s gene enrichment and pathway analyses revealed that many of these variants occurred in genes related to the extracellular matrix and the blood-brain barrier basement membrane, including fibronectin 1 (FN1) and collagen 6 (COL6A2). Because of its presence across all study cohorts, fibronectin 1 rs140926439 variant was prioritized for further investigation. PMID: 38598053.

Data from the Alzheimer’s Disease Sequencing Project (ADSP), Alzheimer’s Disease Genetics Consortium (ADGC), and United Kingdom Biobank (UKB) confirm that the FN1 variant rs140926439 provides protection against APOEε4-related AD risk. This FN1 variant reduces the risk of developing AD by 71% and delays the age of onset in APOEε4/4 individuals by an average of 3.37 years. FN1 expression in the human brain is also influenced by APOE genotype. FN1 accumulates at the blood-brain barrier (BBB) in AD and is expressed most highly in APOEε4/4 brains, followed by APOEε3/4, and then APOEε3/3. Post-mortem analyses of APOEε4/4 brains from donors without AD showed FN1 protein levels similar to those of controls, while APOEε4/4 donors with AD displayed significant FN1 accumulation at the BBB. Additionally, individuals with APOEε4 genetic backgrounds carrying the FN1 variant exhibited brain characteristics similar to those seen in APOEε3 AD cases. These included comparable levels of amyloid aggregation, basement membrane thickening, microglial activation, and astrocyte reactivity, closely matching findings in control cases without AD. Overall, these data suggest that FN1 accumulation is a pathological consequence of APOEε4, and the FN1 loss-of-function variant rs140926439 protects against APOEε4-mediated AD.

To explore how the FN1 genetic variant might protect against the effects of APOEε4 in AD, Dr. Caghan Kizil and Dr. Mayeux's teams conducted studies using zebrafish model of AD, humanized APOEε4 and APP knock-in mouse models as well as isogenic iPSC-derived human cells. In their research, they found that reducing FN1 levels in zebrafish helped improve the brain’s ability to clear amyloid, reduce inflammation, protect synapses, and restore the blood-brain barrier (BBB) function. Building on this, they discovered through single-cell RNA sequencing, functional sufficiency and necessity experiments, and targeted pharmacological treatments in zebrafish that FN1 plays a key role in controlling the health of the BBB. Specifically, FN1 helps regulate a molecule called vascular endothelial growth factor A (VEGFA), which is essential for maintaining the vital interaction between brain cells and blood vessels. This regulation is crucial for the BBB’s integrity in AD. They also found that FN1 buildup, which disrupts these protective pathways, depends on the presence of the APOEε4 gene. In further studies with both mouse models and human cell types engineered to carry either the APOEε3 or APOEε4 genotype, as well as a zebrafish model of amyloidosis, they observed that APOEε4 increased FN1 production and its deposition at the BBB. This harmful accumulation of FN1 was confirmed in brain tissue from AD patients with the APOEε4/4 genotype, where increased FN1 deposits were found around blood vessels compared to healthy controls. Overall, these findings suggest that FN1 is a proximal mediator in APOEε4-related damage to the BBB. FN1’s effects on the BBB are mediated through VEGFA-dependent signaling pathways, and a protective variant of FN1 (rs140926439) appears to counteract APOEε4’s negative effects by preventing FN1 buildup at the BBB. This research offers new insight into how certain genetic factors may protect the brain in AD.

Additional functional studies in mice and zebrafish, along with validation in human tissue, consistently suggested that FN1 regulates BBB integrity through a VEGFA-dependent pathway (unpublished). FN1 appears to act on integrin signaling in astrocytes via the Focal Adhesion Kinase (FAK) mechanism; inhibiting FAK led to an increase in VEGFA expression in astrocytes. Loss of FN1 upregulated VEGFA, altering signaling and expression in endothelial cells. Furthermore, VEGFA inhibition compromised vascular integrity by affecting molecular pathways essential to endothelial and astrocyte homeostasis. The team also confirmed these downstream gene and protein expression changes in BBB cell types identified in the zebrafish model and validated in mouse models through post-mortem analyses of human brains, underscoring the role of the APOE-FN1 signaling axis as a pathological regulator of BBB dysfunction in Alzheimer’s disease.

Since the FN1 variant appears to protect against AD in individuals with at least one APOEε4 allele, developing therapeutic compounds that mimic this variant may offer a promising treatment pathway. Using computational chemistry approaches, Dr. Kizil and Dr. Mayeux's teams identified candidate small molecules that can replicate the protective effects of the FN1 variant. They conducted in silico structure-activity relationship (SAR) studies, followed by in vitro efficacy and safety testing, to evaluate and confirm the effectiveness and safety of these selected candidates. Currently, eight compounds are under evaluation in APOEε4-expressing astrocytes. Two of these compounds selectively bind to FN1, stabilizing it without interacting with active FN1 in blood or brain tissue. One compound specifically reduces FN1 levels in an APOEε4-dependent manner in humanized mouse astrocytes and iPSC-derived isogenic human astrocytes. These findings suggest that targeting fibronectin deposition in the extracellular matrix or enhancing its degradation could represent a viable therapeutic strategy for AD, for individuals at higher risk due to APOEε4.

Discussion

Is there any association between collagen and fibronectin in the context of AD? Does fibronectin have a role outside of blood vessels?

Yes, there is an association between collagen and fibronectin in AD. Both are involved in the extracellular matrix (ECM) and blood-brain barrier (BBB) integrity. Fibronectin interacts with collagen to support ECM structure, which is often disrupted in AD. Outside of blood vessels, fibronectin also influences cellular adhesion, signaling, and repair processes in the brain, impacting astrocyte and neuron interactions and contributing to neuroinflammation and tissue remodeling in AD.

Could FN1 influence CAA or small vessel disease?

The Mayeux laboratory is interested in looking at FN1 in both contexts because fibronectin surrounds blood vessels. In addition, 30 percent of autopsied AD cases have small vessel disease.

Instead of increasing production of plaques, could fibronectin 1 buildup impair clearance of plaques by microglia or astrocytes?

Brain injury via traumatic brain injury or radiation exposure increases fibronectin 1 in the same region of the BBB. Therefore, this buildup may be a reaction to an external process or as a result of amyloid accumulation, and FN1 in turn may prohibit clearance. Amyloid likely appears first and triggers FN1 buildup, which then could impair amyloid clearance.

Are protein expression levels of this FN1 variant and wild type comparable? Does the mutant variant itself protect against APOE4-related AD risk or is protection due to levels of available extracellular matrix proteins?

The Mayeux laboratory is currently investigating these possibilities in animal models. A specific conformational change in FN1, not altered expression level, might be conferring this protective effect.

In LOAD, is FN1-related BBB dysfunction a cause or downstream effect of the amyloid cascade?

Amyloid likely triggers the cascade and FN1 is a downstream effect.

Does the FN1 variant cause any biochemical changes, including solubility?

Based on in silico analysis, the protective variant in FN1 is located in the fibronectin 2 domain. This domain is large and changes the protein’s stiffness and interaction with other ECM proteins. Activation of FN1 requires binding to integrins, and this binding requires a conformational change in FN1 that is impacted by the presence of this variant.

What central nervous system cell types express the FN1 protective variant?

The Mayeux laboratory is currently conducting scRNA-seq to identify cell types that express the variant. Existing data suggests that astrocytes are likely the major central nervous system (CNS) cell type that expresses FN1, but FN1 is also expressed in endothelial cells and vascular smooth muscle.

Interaction of Haptoglobin and APOE in Alzheimer Disease

Jonathan L. Haines, Ph.D., Case Western Reserve University

Haptoglobin (HP) and APOE have roles in shared pathways and can also physically bind in CSF, suggesting that HP may affect APOE-related risk for AD. Haptoglobin scavenges hemoglobin to reduce oxidative stress, as free hemoglobin is prone to auto-oxidation and is a potent pro-oxidant. Oxidative stress in the brain can also impact APOE lipid clearance, and HP itself influences the interaction between APOE and amyloid. HP also alters amyloid uptake in astrocytes. Notably, certain HP alleles are associated with neurocognitive impairment in HIV-positive individuals. Together, these cellular mechanisms suggest relevance for HP in AD.

Two main HP alleles (HP1 and HP2) form distinct protein conformations but are difficult to differentiate in genotyping assays. HP1 alone only forms dimers, while HP2 alone can form trimers, tetramers, pentamers, and hexamers. When both alleles are present, HP1 and HP2 can form complexes ranging from dimers through hexamers. HP’s genetic structure makes it difficult to directly genotype samples using SNP arrays. HP1 and HP2 alleles contain a common copy number variant (CNV) that spans two tandem exons, three and four, of HP; HP1 contains one copy, while HP2 contains two copies of exons three and four. Short-read WGS also cannot accurately genotype HP due to the exon duplication. In addition, review of existing genome-wide association study (GWAS) data of SNPs near HP shows no direct genetic association with AD; the most significant association was located 93 kB downstream. Alternatively, HP alleles can be imputed using a custom reference panel. Imputation with TOPMed and hard calls with a dosage greater than 0.9 accurately called alleles and was verified using unique exon junction expression from RNA-seq data. Using imputation with TOPMed and a dosage cutoff of 0.9 enables researchers to genotype individuals for HP based on existing SNP array and WGS data.

This genotyping method was applied to ADGC data to determine whether HP has an effect on AD risk and whether a genetic interaction between HP2 and APOE can affect AD risk. All logistical regressions adjusted for sex, age, and the top three principal genetic components. No main effect for HP was detected, but the HP2 allele appears to attenuate the effect of APOE on AD risk in certain APOE genotypes in European ancestral backgrounds. Specifically, HP2 attenuates the effects of APOE2 and APOE4 in APOE2/E2, APOE2/E3, and APOE2/E4 genetic backgrounds. HP2 does not attenuate effects in APOE3/E3 or APOE3/E4 backgrounds and may actually slightly increase AD risk in APOE4/E4. Similar attenuations were observed in individuals of African descent.

Discussion

Does the interaction between HP2 and APOE background impact biomarker levels in humans?

Biomarker data are currently being collated from ADGC and other datasets to assess potential differential biomarker levels.

APOE ε2 Allele and Protective Variants in APOE ε4/ε4 Carriers on Alzheimer’s Disease Risk

Gyungah Jun, Ph.D., Boston University

By analyzing observational data, including genetic variants, in the context of AD, researchers can leverage the randomization of these variants to identify potentially druggable pathways that, when perturbed, mimic the protective effects of these variants. Because APOE2 is generally protective against AD, genetic variants detected in APOE2 AD cases but not APOE2 controls may disrupt the APOE2 protective mechanism. Because APOE4 is generally a risk allele for AD, genetic variants detected in APOE4/E4 controls but not in APOE4/E4 AD cases may provide protection against the APOE4/E4 genotype. This strategy targets APOE backgrounds not previously leveraged for detecting protective or risk genes, as most of the previous AD risk gene findings originated from APOE3/E3 and APOE3/E4 data.

In APOE2 carriers, GWAS data yielded near significance for protein phosphatase 2 catalytic subunit beta (PPP2CB) as a risk modulator. The A allele at rs11729832 was associated with higher PPP2CB expression compared to the G allele. PPP2CB encodes the C subunit of protein phosphatase 2A (PP2A), which dephosphorylates tau. In brain tissue, PPP2CB protein level strongly correlates with the classical complement component C4B expression in APOE2/E3 and APOE3/E4 genotypes. Furthermore, PPP2CB protein expression in brain tissue strongly correlates with Aβ42 levels in APOE3/E4, but not in APOE2/E3. A transcriptome-wide association study in APOE2 carrier control and AD brains identified differentially expressed genes, including C4A, C4B, and glial fibrillary acidic protein (GFAP). C4A and C4B were expressed at higher levels in AD brains compared to control brains, but this differential expression was unique to APOE2 carriers. Analysis of co-expression networks determined that an APOE2-related co-expressed gene network containing C4A and C4B also included genes previously implicated in AD, including complement receptor 1 (CR1) and HBEGF. Together, these data suggest that an APOE2-related protective mechanism may be linked to classical complement components C4A and C4B as well as GFAP.

In individuals with APOE4/E4 genotype, GWAS analysis identified four candidate risk modulator SNPs (unpublished). The top protective SNP was rs75983775, which is located near tribbles pseudokinase 2 (TRIB2). T allele carriers from the Framingham Heart Study at this SNP had higher TRIB2 expression and higher cognitive scores compared to non-T allele carriers. The association between the T allele and TRIB2 expression was confirmed by comparing TRIB2 expression in APOE2/E2 and APOE4/E4 neurons and astrocytes. TRIB2 is a pseudo serine/threonine kinase that can interact with E3 ubiquitin ligases to control protein stability of downstream effectors. TRIB2 is also a scaffold enzyme that integrates and modulates signals flowing through MAPK and AKT modules. TRIB2 is also involved in regulation of homeostatic cytokines that contribute to loss of naïve T cells and antibodies to TRIB2 may contribute to neuronal death. These results demonstrate that rs75983775 may exert a protective effect in the APOE4/E4 background via a T cell immune response.

The APOE4 allele also appears to modulate epigenetic profiles of genes near and within the APOE locus. Compared to non-carriers, APOE4 carrier methylation data indicate hypermethylation at CpG sites in the APOE locus in both blood and brain tissue. In addition, APOE4 carriers exhibit hypomethylation compared to non-carriers at CpG sites in the APOC1 locus in blood but not brain tissue. Lastly, a CpG site in TOMM40 is hypomethylated in APOE4 carriers compared to non-carriers in brain tissue but not blood. Together, these data indicate that epigenetics require further consideration in APOE studies, particularly in a cell type-dependent manner.

Discussion

Is the mechanism for PPP2CB modulation of APOE-related AD risk related to the previously identified risk variant in PPP2A that modulates levels of both Aβ and tau in the CSF?

PPP2CB and PPP2CA are likely in the same pathway, and while PPP2CB expression is highly correlated with PPP2A expression in the brain, these expression levels are not APOE background-dependent. A larger sample size is needed to compare the PPP2CB and PPP2CA variants.

How might the complement cascade contribute to tauopathy?

The C4 haplotype is located in major histocompatibility complex (MHC) 3, between MHC1 and MHC2. Prior studies have linked the C4 haplotype to neurodevelopment and neurodegeneration. C4 and MHC3 may impact risk together with MHC2, and the C4 haplotype is related to CNVs in MHC.

Which cohorts were used to screen for APOE2 modifiers?

ADGC data was used for the initial screen, and the Jun laboratory will screen ADSP as well to increase sample size and diversity; this is especially important given that APOE2 carrier AD cases are very rare.

What cell types express TRIB2 in APOE2 carriers?

Based on iPSC data, TRIB2 is expressed in astrocytes and neurons, but its expression in microglia has not yet been investigated.

Rare Protective APOE Variants

Michael Greicius, M.D., M.P.H., Stanford University

Examination of unique human case studies involving APOE loss-of-function variants can provide insight into whether the APOE4 variant is inherently toxic or has reduced function compared to APOE3. A case report of a patient homozygous for a very rare frameshift/stop-gain mutation in APOE indicated that this person had no measurable APOE in lipoprotein fractions. This patient exhibited severe cardiovascular disease and xanthomas. Although the patient’s CSF levels of amyloid and tau were normal at age 40, their memory testing results were abnormal.

Dr. Michael Greicius hypothesized that if APOE4 has reduced function compared to APOE3, then loss of a single APOE4 copy in an APOE3 individual should increase AD risk, while if APOE4 is toxic, loss of a single APOE4 copy in an APOE4 individual should reduce AD risk. After screening 56,000 exomes/genomes from ADSP, Dr. Greicius identified seven individuals with loss-of-function variants on APOE (4 single-nucleotide variants, including one seen in 3 individuals, and one deletion encompassing exons 1 and 2). Of these seven carriers, five were controls and only one was an AD case (the seventh carrier has a diagnosis of PSP).

The sole AD case identified in the ADSP dataset had an APOE3/E4 genotype with an early stop-gain in the APOE3 allele, resulting in an effective APOE4/null individual. This individual exhibited onset of AD symptoms at age 75 and died at age 87. Autopsy confirmed that brain pathology was consistent with an AD diagnosis. The mean age of onset of AD symptoms in APOE4 homozygotes is 68.4 years and 75.5 years in APOE4 heterozygotes. Relative to these typical ages of onset, the loss of the APOE3 allele in this case report did not seem to change AD risk much in comparison to APOE4 heterozygotes.

One healthy older control identified in the ADSP dataset had an APOE3/E4 genotype with a stop-gain in the APOE4 allele, resulting in an effective APOE3/null individual. This individual had amyloid-negative CSF at age 76 and was still cognitively normal at age 79. Typically, two-thirds of people with no clinical symptoms and the APOE3/E4 genotype are amyloid positive at age 76. This suggests potentially protective status for this healthy older control, as many APOE3/E4 individuals already are amyloid positive at 76, not to mention the fact that many APOE3/E4 individuals are no longer cognitively healthy at 76 but may have MCI or AD.

Another healthy older control identified in the ADSP dataset had an APOE3/E4 genotype with a stop-gain in the APOE4 allele, resulting in an effective APOE3/null individual. After death at age 90, this individual’s autopsy was notable for the absence of amyloid plaques in the parenchyma and blood vessels. This finding is very uncommon in those with the APOE3/E4 genotype. Of the 1,700 individuals in the National Alzheimer’s Coordinating Center (NACC) database who are APOE3/E4, exceedingly few had a complete absence of amyloid in parenchyma and very few had complete absence of CAA. Together, this information suggests that loss of APOE4 in this individual was highly protective against AD.

Based on these case reports, APOE4 likely has a toxic GOF, supporting a potential therapeutic strategy that knocks down APOE4 in APOE4/E4 and APOE3/E4 individuals. Because seven individuals with loss of APOE4 were long-lived, this strategy should be relatively safe. Further characterization of missense variants may provide insight into a potential small molecule approach that mimics the variants’ protective effects. For example, V236E and R251G variants are associated with decreased AD risk across multiple cohorts. Individuals with APOE3/E3 V236E or APOE3/E4 R251G genotypes have an AD risk equivalent to those with an APOE2/E3 genotype. V236E is near and R251G within the lipid-binding domain of APOE, suggesting that targeting this domain may be a viable therapeutic approach. Indeed, the CS6253 peptide (Artery Therapeutics, Inc.) is a peptide modeled on the C terminus of APOE including amino acid 251 and is an agonist of ABCA1. Artery Therapeutics has recently completed a Phase I safety clinical trial for this peptide.

Discussion

How does single copy loss of APOE compare to APOE3/E3 in terms of AD risk and pathology?

Even when comparing the APOE3/null 90 year-old control to APOE3/E3 subjects in the NACC autopsy database, it is uncommon to find APOE3/E3 subjects at age 90 who have no amyloid plaque in the parenchyma or the blood vessels. Are the expression levels of other apolipoproteins higher in APOE null patients as a compensatory mechanism?

Dr. Greicius has not yet analyzed apolipoprotein levels in these null cases so it remains possible that the remaining normal APOE allele increases its expression to compensate for the loss-of-function allele.

Do individuals with protective variants in APOE, like R251G, exhibit differences in autopsy or biomarker data? These variants are rare and, thus far, Dr. Greicius has not found carriers with biomarker or autopsy data. How does this analysis handle selection bias in terms of survival effects due to midlife cardiovascular disease? Are there data on plasma lipid levels in patients with at least one APOE3 copy missing?

These individuals with loss of one copy of APOE3 are extremely rare, which limits stratified analyses. Dr. Greicius identified another APOE3/null individual in UKB—a female with unspecified dementia in the age range of 70 to 79 with several additional dementia diagnoses added at age 79. However, UKB lacks detailed phenotypic data including biomarkers, limiting any further analyses.

What are the lipid levels for individuals missing a single copy of APOE3?

People in the UKB with one copy of a loss-of-function variant appeared to have normal lipid levels, however, many UKB participants are already on statins, complicating efforts to determine if an APOE loss-of-function variant affects lipid levels.

Do these APOE3 variants impact protein stability?

Dr. Greicius is working with a protein chemist to characterize the effects of these variants on protein stability and physical interactions between ABCA1 and APOE.

Arginine in the C terminus of APOE can impact APOE cleavage, and therefore, heparin sulfate proteoglycan (HSPG) binding. Do any of these variants impact protease cleavage or HSPG binding?

Dr. Greicius is working with a protein chemist to understand the impacts of these variants on APOE-HSPG binding. Notably, the relationship between HSPG binding and AD risk is likely quite complex. For example, the R145C variant increases AD risk and has the same impact on HSPG binding as the Christchurch variant that is hypothesized to decrease AD risk.

Do these data indicate that reducing APOE by 50 percent in the brain causes no adverse effects?

Based on the phenotypes reported in the loss-of-function paper, this amount of reduction appears safe. It is possible, though, that these loss-of-function variant carriers have increased expression from the normal APOE allele to compensate for the loss-of-function allele.

APOE-Genotype Dependent Single-Cell Transcriptomics of Alzheimer’s Disease

Li Hui Tsai, Ph.D., Massachusetts Institute of Technology

The diverse roles of APOE suggest that APOE4 may have cell type-specific effects in the human brain. Using scRNA-seq on 12 APOE3/E3, 12 APOE3/E4, and 8 APOE4/E4 samples from the Religious Orders Study Memory Aging Project (ROSMAP) identified differentially expressed genes in specific cell types. APOE4 altered several cellular pathways in a cell-type-specific manner. Multiple processes were globally upregulated in the brain, including Aβ formation, NF-κB signaling, and cholesterol processing in APOE4/E4 cells. Overall, the differentially regulated processes identified in this experiment have been previously implicated in APOE biology.

Because APOE binds lipids, APOE4 may have cell type-specific effects on cholesterol biosynthesis. Based on scRNA-seq data, cholesterol biosynthesis is upregulated in APOE4/E4 oligodendrocytes and slightly downregulated in APOE4/E4 astrocytes. This effect in oligodendrocytes is an APOE4 dose-dependent effect. BODIPY staining of APOE4/E4 oligodendrocytes further confirmed the transcriptomic finding of upregulated cholesterol biosynthesis. Dysregulated cholesterol synthesis is also associated with reduced myelination in APOE4 carriers compared to non-carriers. Treatment of APOE4/E4 oligodendrocytes with 2-hydroxypropyl-β-cyclodextrin to increase cholesterol efflux and lower cellular cholesterol restored axonal myelination in APOE4/E4 co-culture and mice.

Furthermore, APOE4 can induce lipid droplet accumulation in microglia, resulting in microglial defects in sensing neuronal activity and neuronal surveillance functions. The bidirectional manipulation of lipid content (i.e., oleic acid to increase lipid burden and long-chain acyl-CoA synthetase 1 (ACSL1) inhibitor to decrease lipid burden) can reversibly drive neuronal surveillance of iPSC-derived microglia. Treatment of APOE3/E3 iPSC-derived microglia with oleic acid increases lipid burden and worsens microglia function, while treatment of APOE4/E4 iPSC-derived microglia with ACSL1 inhibitor reduces lipid burden and improves microglia function.

Further single cell transcriptomic interrogation of 427 prefrontal cortical samples from ROSMAP cohorts with a wide range of AD pathology and cognitive impairment identified multiple biological processes perturbed in more severe AD cases. Perturbed processes included mitochondrial function, mRNA metabolism, synaptic signaling, cohesion complex expression, and lipid metabolism. The dysregulation of lipid metabolism seen in cells from donors with severe AD mirrors the effects seen in APOE4/E4 astrocytes and oligodendrocytes. Therefore, loss of lipid homeostasis likely constitutes an important pathogenic step in AD.

Discussion

Are these myelination defects due to cell autonomous expression of APOE in oligodendrocytes?

Microglia clearly play an important role in myelination. Current experimental results cannot rule out secretion of APOE from other cell types and subsequent APOE binding to receptors on oligodendrocytes. APOE likely functions by binding lipid cholesterol as a secreted protein, even in monocultures.

Why do APOE4 oligodendrocytes have increased cholesterol but less lipidated APOE?

APOE4 exhibits reduced lipid binding regardless of intracellular levels of cholesterol. Due to reduced lipid binding of APOE4, cholesterol biosynthesis is dysregulated, resulting in increased intracellular accumulation.

Are there sex differences in the lipid dysregulation in APOE4 cells?

Current experiments lack sufficient statistical power to identify any sex effects. However, larger datasets will enable these comparative studies.

What is the order of the events observed in different APOE4 cell types?

Current data is insufficient to predict the specific order and causality of these events.

The role of APOE genotype in microglial behavior and gene expression

Alison Goate, D.Phil., Mount Sinai

In microglia, background haplotypes but not APOE isoforms drive major gene expression differences. Transcriptomic analysis of microglia determined that although APOE3 and APOE4 isoforms in an identical genetic background had similar expression patterns, the same isoform, either APOE3 or APOE4 in different genetic backgrounds had significantly different transcriptomes. Notably, microglia with the APOE4 isoform and APOE4 genetic background as well as microglia with the APOE4 isoform and APOE3 genetic background had increased lipid droplet organization and reduced secondary lysosome activity, regulation of cholesterol efflux, and eukaryotic translation initiation. Together, these results suggest that haplotype likely has important effects on APOE function in microglia.

Using cellular and functional assays, Dr. Alison Goate further dissected the effects of APOE haplotype on microglial function. Compared to iPSC-derived microglia containing the E4 SNP in an APOE3/E3 haplotype, microglia with the APOE4 SNP in an APOE4/E4 haplotype exhibited no change in phagocytosis of myelin fragments, lysosome acidification, and lysosome proteolysis, but reduced lysosomal mass, APOE expression, and baseline protein translation. Together, these results suggest a haplotype rather than isoform effect. Conversely, the level of BODIPY-positive lipid droplets remained unchanged in APOE4/E4 versus APOE3/E3 population lines but was increased in microglia with the E4 SNP in an APOE3/E3 background compared to the E3 SNP in an APOE3/E3 background. This result suggests a protein isoform-specific effect on BODIPY positive lipids.

scRNA-seq data indicate isogenic microglia expressing the APOE4 isoform have a reduced fraction of DAMs and negative enrichment for both DAM and lipid-associated microglia marker genes compared to the APOE3 isoform (unpublished). Pharmacological activation of LXR signaling rescued the transcriptomic changes observed in APOE4 isoform microglia affecting lysosome processing and lipid metabolism.

Discussion

Did these experiments use iPSC lines from healthy individuals or those with AD?

These experiments used a mixture of donors with and without AD. Analyses used AD status as a covariate, but during the process of generating iPSCs, epigenetic differences that would impact iPSC-derived microglia are likely erased.

Could the APOE variant impact the differentiation of iPSC-derived microglia?

The Goate laboratory has not observed any gross differences during the differentiation protocol, although that observation does not preclude differences in differentiation.

Could some of the phenotypic differences between APOE3/E3 and APOE4/E4 be explained by different APOE expression levels?

These phenotypic differences could be due to differences in APOE expression level, but distinguishing expression level effects from APOE background effects is difficult in human tissue.

Why was there a difference in lysosomal mass but not phagocytosis?

The lack of statistical significance for phagocytosis may be due to a small sample size.

Does applying a challenge to these cell lines result in differential effects between APOE3/E3 and APOE4/E4?

The Goate laboratory has applied myelin challenges to these cells, but not lipopolysaccharide or amyloid challenges.

The role of APOE genetics in glial lipid metabolism and inflammation

Julia TCW, Ph.D., Boston University

Lipidosis associated with APOE in glia is a hallmark of AD. In AD patient brains, astrocytes appear reactive, and in AD mouse models, microglia are packed around amyloid plaques with APOE clustering with these microglia. The APOE4/E4 genotype exhibits brain cell type-dependent effects seen across iPSC, postmortem AD brain, and mouse experimental models.

Matrisome gene sets—including extracellular matrix genes, core matrisome and matrisome-associated genes, and cytokine genes—are enriched in AD brain APOE4/E4 astrocytes (TCW et al. Cell 2022 PMID: 35750033). Potential AD-related functions associated with the matrisome include increased chemotaxis of phagocytes, increased activation and inflammation, and increased lipid synthesis. These transcriptional changes appear specific to humans, especially lipid metabolism. Lipid metabolism pathways are also mainly enriched in human APOE4/E4 microglia and astrocytes, including increased cholesterol synthesis and accumulation as well as reduced lipid catabolism and cholesterol efflux. This lysosomal cholesterol accumulation is caused by de novo cholesterol biosynthesis in APOE4/E4 astrocytes, possibly because the endoplasmic reticulum cannot sense existing cholesterol. These excess lipids co-localize with lysosomes.

Consistent with lipid accumulation in lysosomes in APOE4/E4 astrocytes, scRNA-seq data indicated that lysosomal genes were among the top downregulated pathways in APOE4 AD astrocytes and human brain tissue compared to non-APOE4 AD astrocytes and human brain tissue, respectively (unpublished). Other downregulated pathways included regulation of autophagy, macroautophagy, and lysosomal protein catabolic process. Numerous degradation pathways converge on the lysosome including macroautophagy, chaperone-mediated autophagy, endosomal microautophagy, endocytosis, and phagocytosis. Notably, lipophagy occurs via the macroautophagy pathway. Using radio-labeled proteins as a substrate for lysosomes in a pulse-chase experiment showed reduced lysosomal and macroautophagic proteolysis in APOE4/E4 compared to APOE3/E3 astrocytes. In addition, free cholesterol and neutral lipid droplet flux was reduced in APOE4/E4 compared to APOE3/E3 astrocytes, suggesting that impaired lysosomal catabolism may contribute to lipid accumulation in APOE4/E4 astrocytes. Multiple autophagy activation methods, including standard serum deprivation, rescued impaired lysosomal and macroautophagic proteolysis in APOE4/E4 astrocytes. This activation also partially ameliorated intracellular lipid accumulation, significantly reduced proinflammatory cytokine and chemokine release, and increased certain anti-inflammatory factors like interleukin 2.

Discussion

Is the disturbance in degradation relevant to lysosomal autophagy in APOE4 carriers?

ABCA1 protein levels are down, and based on cholesterol chasing experiments, secretion and efflux are downregulated. Therefore, these cells are synthesizing cholesterol but cannot secrete it properly. Perhaps other factors contribute to this dysregulation, such as lower APOE and ABCA1 expression as well as missing efflux machinery. This could explain why activation of LXR increases APOE levels, but also reverse phenotypes seen in glia cells.

Given that both free cholesterol and neutral lipids accumulate intracellularly, could APOE4 affect acid lipase and NPC1 and NPC2 machinery?

Future lipidomic assays should provide further insight into this question. NPC1 is reduced in lysosomes in AD cases, but in Niemann Pick Disease, researchers report that phenotypes are due to post-translational modifications of NPC1 and NPC2. Even though phenotypes seen in AD and Niemann Pick Disease are similar, experimental results suggest different mechanisms.

Why are accumulation of cholesterol and other lipids in lysosomes and increased cholesterol biosynthesis occurring in APOE4 cells? Did the gene ontology analysis include cholesteryl ester transferase—an enzyme involved in cholesterol biosynthesis and degradation—under the cholesterol biosynthesis pathway?

While the level of lipid sensor such as SCAP is increased, the lipid synthesis proteins, cleaved SREBP2 and HMGCR, HMG-coA reductase are increased in APOE E4/E4, it seems that it cannot detect the intracellular level of cholesterol as the lipids are accumulated in lysosome but not export out from lysosome to ER in part due to low level of cholesterol binding lipid exporter like NPC1. APOE4/E4 AD cases seem to exhibit impaired cholesterol degradation, which is why future work will focus on lysosomal processes. Glia cells try to digest these lipids through multiple pathways that may be impaired.

In LOAD APOE4 carriers, is APOE4-mediated impairment in amyloid clearance or uptake contributing more to AD pathogenesis?

Degradation appears to be the main contributor to AD pathogenesis in these models. Some uptake issues have been observed, but not as drastic as degradation phenotypes (unpublished).

The role of genetic variants and their influence on the immune response in myeloid cells/microglia

Christopher K. Glass, M.D., Ph.D., University of California, San Diego

Prior GWAS have found that most common AD risk alleles reside in noncoding regions of the genome, which creates challenges for determining which genes and cell types are impacted by particular variants. To address this, the Glass laboratory has mapped active enhancers and promoters in neurons, astrocytes, oligodendrocytes, and microglia in the human brain by combining fluorescent activated nuclear sorting with epigenetic assays associated with open and active chromatin. This method has identified tens of thousands of putative enhancers specific to each cell type in the brain. Based on linkage disequilibrium score regression analysis, risk alleles most associated with psychiatric traits were most strongly enriched in neuronal enhancers, whereas risk alleles for AD were most strongly enriched in microglia enhancers. These data support the concept that AD risk alleles exert phenotypic effects by altering expression in microglia.

These dynamic enhancer landscapes can be used to decode microglia phenotypes and mechanisms in healthy and disease states. Enhancers play critical roles in mediating changes in cellular responses to external signals. Enhancers integrate these signals by providing sites of action of specific combinations of cell-specific and signal-dependent transcription factors. Changes in cell state or fate typically involve changes in activities of hundreds to thousands of enhancers. After identifying a set of enhancers that become activated in response to a stimulus, subsequent analyses can identify sequence motifs and their corresponding transcription factors that are driving cell fate and state changes.

This enhancer and motif analysis method was used to characterize the phenotypic changes in microglia and the transcription factors driving them during tau-dependent neurodegeneration. This enhancer and motif analysis was applied to scRNA-seq data from TE4 and APOE4 mouse-derived microglia. TE4 mice exhibit severe, microglia-dependent neurodegeneration compared to PS19 APOE3. scRNA-seq data from hippocampi isolated from APOE4-only mice and TE4 mice were analyzed to identify differentially active enhancers specific to this neurodegeneration phenotype. TE4 microglia expressed several hundred genes not expressed in microglia associated with amyloid plaques. These microglia also exhibited marked cytoplasmic and lysosomal lipid accumulation. These microglia with these expression patterns and phenotypes were deemed tau-APOE4-responsive microglia (TERM).

To determine whether TERMs were anatomically or functionally related to neurodegeneration observed in TE4 mice, the Glass laboratory designed a probe set and used multiplexed error-robust fluorescent in situ hybridization (MERFISH) to interrogate the locations of molecularly distinct microglia phenotypes in mouse brains from APOE4 mice, TE4 mice, and APP/PSEN1 mice. This approach identified specific subsets of microglia exhibiting molecular phenotypes consistent with homeostatic microglia, amyloid-responsive microglia, and TERM. Homeostatic microglia were distributed throughout brains of all three models but were underrepresented in hippocampal regions of the TE4 brain. TERMs were nearly absent from APP/PSEN1 and APOE4-only mouse brains but abundant in TE4 hippocampi in proximity to areas of neurodegeneration. This pattern of expression for TERMs was similar to glycoprotein nonmetastatic melanoma protein B (GPNMB) expression, a protein previously implicated in activated microglia in AD. Together, these results suggest that TERMs either drive or respond to neurodegeneration in TE4 mice.

To identify active enhancer regions and associated transcription factors specific to TERM activation, the Glass laboratory used ATAC-seq analysis on APOE4-only mouse microglia (mainly homeostatic) and TE4 mouse microglia (mainly TERM). This analysis identified over 900 regions in the genome with at least a two-fold increase in open chromatin in TE4 compared to APOE4-only microglia. Some of these open regions were identified near lipoprotein lipase (LPL), secreted phosphoprotein 1 (SPP1), and basic helix-loop-helix family member e40 (BHLHE40) genes, all of which were also transcriptionally upregulated in TERMs. Motif enrichment analysis identified DNA recognition sequences for CCAAT/enhancer binding protein (C/EBP), PU.1, activator protein 1 (AP-1)/activating transcription factor 3 (ATF3), microphthalmia-associated transcription factor (MITF)/basic helix-loop-helix (BHLH), and early growth response 2 (EGR2) transcription factors. Ultimately, these data support the inference that environmental signals in the context of interaction between tau and APOE4 lead to alteration of enhancer landscapes of microglia, enabling certain transcription factors to drive the TERM phenotype. Additional functional studies are currently confirming the roles of these transcription factors in driving specific aspects of microglia phenotypes.

Discussion

Could lipids mediate the described mechanism?

Lipids are accumulating in the lysosomes of TERMs. Other work in the Glass laboratory is using similar scRNA-seq and ATAC-seq to study primary lysosomal storage diseases, including mucopolysaccharidosis 3A, caused by loss of N-sulfoglucosamine sulfohydrolase (SGSH). SGSH knockout-derived cells had overlapping transcriptional and epigenetic programs to TE4 cells, indicating that lysosomal dysfunction is likely a partial driver of the molecular phenotype seen in TERMs. Approximately 1,000 genes were dysregulated in SGSH knockout mice at 8 months of age, with a 30 percent overlap with genes upregulated in TERMs. Some phenotypes of TERMs are also related to lysosomal dysfunction and may be driven by the accumulation of lipids.

How much do the transcriptomes and enhancers of TERM and DAM overlap?

Approximately one-third of dysregulated genes are overlapping for TERM and DAM, another one-third biased toward DAMs, and the final third biased toward TERMs. Genes associated with both DAMs and TERMs were enriched for immune genes, including those involved in TNF pathways as well as cytokines. Genes most associated with DAMs were highly enriched for lysosomal function, oxidative phosphorylation, and translation. Genes associated with TERMs indicate an immune signature, but the enrichment for immune-related genes was not as strong as the enrichment of genes associated with both TERMs and DAMs.

Enhancer differences follow a similar pattern to transcription differences. Motif enrichment analysis identified a similar set of motifs across all three groups of enhancers (i.e., overlapping, DAM-biased, and TERM-biased). Therefore, microglia may be using a common palette of signal-dependent transcription factors in different combinations to drive overlapping and disease-specific (DAM versus TERM) phenotypes.

A previous publication from the Butovsky laboratory detected upregulation of genes associated with reactive microglia in both APP1/PSEN1 and superoxide dismutase 1 (SOD1) mouse models. Were TERMs not detected in these more recent APP1/PSEN1 experiments due to lower disease severity? Would older mice in this experimental paradigm have TERMs?

DAM and TERM could represent two completely independent phenotypes or two different points on a continuum of severity of a single phenotype. TERMs are only seen in TE4, not APP1/PSEN1, but some genes are upregulated in both models including C-type lectin domain containing 7a (CLEC7A) and cd9. However, DAMs are seen in both TE4 and APP/PSEN1 models. Some aspects of DAMs and TERMs may be along a single severity continuum, but because some TERM and DAM spatial localizations differ, some aspects may be distinct. Currently, the Glass laboratory is investigating triggering receptor expressed on myeloid cells 2 (TREM2) signaling to determine whether this pathway can explain the divergence of open chromatin in DAMs and TERMs.

Are the inflammatory responses associated with amyloidosis and tauopathy overlapping or completely independent?

Based on the experimental models used in the Glass laboratory, the response to amyloid pathology may not necessarily be a direct response to amyloid. Because DAMs are present in the TE4 model that does not have amyloid, DAM activation may be a response to a pathology downstream of amyloid deposition. DAMs in TE4 mice are diffuse and not specific to regions with neurodegeneration, so the DAM phenotype may be related to a certain tissue injury common to both TE4 and APP/PSEN1 models not associated with cell death. In contrast, TERMs are only observed in the context of neurodegeneration and cell death.

The Glass laboratory is now trying to assess whether TERM and DAM phenotypes are present in AD patients. By using gene signatures that distinguish TERMs from DAMs, myeloid cells from deeply phenotyped AD patients could be characterized as either TERM-like or DAM-like. In addition, overlapping signatures for primary lysosomal disorders can provide insight into any commonalities between those disorders and AD.

An allelic series of lipidated ApoE drives CNS lipofuscinosis (unpublished)

Gilbert Di Paolo, Ph.D., Denali Therapeutics

APOE2 and APOE3Ch protective variants are associated with decreased interactions with membrane receptors LDLR and HSPG. In contrast, APOE4 exhibits increased membrane receptor binding along with other attributes that likely contribute to pathogenicity. However, whether reducing APOE interactions with membrane receptors is generally protective remains unclear.

APOE binds a variety of lipids, and certain lipid cargoes may contribute to APOE-related AD pathology. Cholesterol esters (CEs) are disease associated with increased levels detected in brains of amyloid and tau AD mouse models, vulnerable brain regions of LOAD patients, iPSC-derived neurons from AD donors, CSF from AD donors, TREM2 knockout iPSC-derived microglia and in vivo, and APOE4/E4 iPSC-derived astrocytes. Together, these data suggest that CEs are important in AD pathogenesis. Downregulation of CEs with acetyl-CoA acetyltransferase (ACAT) inhibitor is protective in amyloid models, while downregulation of polyunsaturated fatty acid (PUFA)-CE with an LXR agonist is protective in an APOE4 tau mouse model. Lastly, brain CE levels correlated with plasma neurofilament light-chain (NfL) in an APOE4 tau mouse model.

CEs may also be associated with lipofuscin—autofluorescent cellular waste consisting of highly oxidized and cross-linked macromolecules, including lipids. Lipofuscin is predominantly found in lysosomes of neurons in aged brains and is a general hallmark of aging. However, the mechanisms of lipofuscin formation are unclear, although factors like oxidative stress, impaired lysosomal degradation, and peroxidation of PUFA-containing lipids might be involved. Lipofuscin also has disease relevance with neuronal ceroid lipofuscinosis, a class of neurodegenerative diseases characterized by lipofuscin deposition. In addition, long and poorly myelinated cortical neurons with higher lipofuscin burden are the most vulnerable to neurofibrillary tangle formation, and neurons free from lipofuscin are generally more resistant to tau pathology in AD.

APOE isoforms differentially induce lipofuscin in iPSC-derived neurons. Expression of the PUFA CE(20:4)/phosphatidylcholine (POPC)-lipidated APOE was sufficient to drive lipofuscinosis. APOE4-expressing neurons accumulated the most lipofuscin, followed by APOE3 and then APOE2. The K146E mutation in APOE4 reduced APOE4-driven lipofuscinosis down to APOE2 levels. This lipidated APOE-induced lipofuscinosis is LDLR-dependent. Blocking LDLR-APOE or replacing the lipid with a peroxidation-resistant CE abolished lipofuscinosis. Notably, application of the antioxidant vitamin E also reduced lipofuscinosis.

APOE4 expression also enhances lipofuscinosis in in vivo mouse brains. In TE4 mice, lipofuscin was increased in the hippocampus and entorhinal cortex, which is where the most severe neurodegeneration occurs in this model. APOE4 expression is associated with even higher lipofuscinosis in these same regions in PS19 mice. This lipofuscin is mainly neuronal but also present in microglia. Strikingly, CE(20:4)/POPC-lipidated APOE4 injection into a mouse brain was sufficient to increase lipofuscin, even in the absence of AD pathology seen in AD mouse models. Consistent with other studies that report an association of AD pathology with endolysosomal dysfunction, a higher lipofuscin load promoted endolysosomal accumulation of tau pre-formed fibrils. Lastly, the Ch variant reduces lipofuscin in APOE3 and APOE4, with a more dramatic reduction in APOE4.

Chronic dosing of LXR agonist GW3965 was neuroprotective and decreased PUFA-CE species and lipofuscin in TE4 mouse brains. Over the course of 3.5 months, chronic LXR agonist selectively decreased brain sterols, especially CE(18:2), CE(20:4), CE(22:5), and CE(22:6), suggesting these sterols may mediate lipofuscinosis.

Discussion

What are the effects of APOE interaction with even more unsaturated fatty acids?

These unsaturated fatty acids are being synthesized and will likely yield results similar to CE(20:4)-APOE.

Do previous reports of filipin staining for unesterified cholesterol in APOE4 astrocytes have any relation to lipofuscin?

Dr. Di Paulo reported no observed correlation between filipin staining and lipofuscin.

Is lipofuscin associated with neurofibrillary tangles or tau?

This association was previously reported in mice and humans. In Dr. Di Paulo’s tau APOE4 models, both the hippocampus and entorhinal cortex exhibit robust pathology. In vitro data suggest that lipofuscin and tau may co-localize in neurons, but in vivo experiments are required to confirm this.

What differentiates ubiquitous lipofuscin observed in aging brains generally and lipofuscin in AD brains?

Lipofuscin is primarily a manifestation of brain aging. Data have not yet been collected on where lipofuscin deposits are located relative to plaques in human and mouse AD brains. The current hypothesis is that while there is an age-dependent accumulation of lipofuscin, Aβ itself is a factor that drives more of the CE increase in AD.

Given that mutations in multiple genes can cause neuronal ceroid lipofuscinosis (NCL), is there any connection between APOE4 and NCL?

Dr. Di Paulo has characterized lipofuscin in granulin loss-of-function models and has seen high levels of lipofuscin, albeit mostly in microglia. Lipofuscin in this model colocalizes with TERMs, as indicated by GPNMB staining. However, whether this buildup mechanism is similar to the buildup seen in AD neurons remains unknown. Interestingly, ceroid-lipofuscinosis, neuronal 5 (CLN5)—a previously reported AD risk factor—is involved in bismonoacylglycerophosphate phospholipid synthesis, which is critical for lysosomal function. Therefore, investigations between lipofuscin and CLN5 in the context of AD are currently being pursued.

APOE4/4 is linked to damaging lipid droplets in Alzheimer’s disease microglia

Michael Haney, Ph.D., University of Pennsylvania

Lipid droplet accumulating microglia (LDAM), initially identified in aged microglia, are dysfunctional and proinflammatory. Lipid droplets are large cytosolic storage vesicles of lipid, mostly filled with triglycerides and sterol esters, that bud off from the endoplasmic reticulum. They can play a variety of roles in various cell types, and dysfunctional lipid metabolism is a major contributor to AD risk.

To better understand lipid droplet formation in the context of AD, scRNA-seq was used to compare human brains from AD cases (APOE4/E4 and APOE3/E3 genotypes) and control APOE3/E3 brains. ACSL1 was upregulated in AD cases, especially in APOE4/E4 brains and is involved in lipid droplet formation. ACSL1 inhibition prevents droplet formation and, in some circumstances, can drive lipid droplet formation. ACSL1 upregulation was specific to microglia, and ACSL1-positive microglia were commonly associated with plaques, suggesting that plaques could be inducing ACSL1 expression.

Results from isogenic APOE4/E4 and APOE3/E3 iPSC-derived microglia were similar to findings from human brains, supporting the use of iPSC-derived microglia as an experimental model. APOE4/E4 iPSC-derived microglia had more lipid droplets compared to APOE3/E3 microglia at baseline and when challenged with fibrillar Aβ (fAβ) aggregates. Lipid droplet-associated genes, PLIN2 and ACSL1 were upregulated in response to fAβ challenge, and inhibition of ACSL1 in this challenge context reduced lipid droplet formation. ACSL1 upregulation is not unique to fAβ challenge—other studies have reported ACSL1 upregulation in microglia in response to lipopolysaccharide (LPS). In the case of fAβ, Dr. Michael Haney posits that de novo lipid synthesis occurs based on deuterated glucose and targeted lipidomics showing glucose incorporation into lipid droplets.

Human LDAMs can have profound effects on neurons in vitro. LDAMs exhibit heightened chemokine and inflammatory cytokine production and defects in phagocytosis and lysosomal accumulation. Conditioned media from APOE4/E4 LDAMs was sufficient to induce phosphorylated tau (pTau) accumulation in iPSC-derived neurons, and this accumulation was APOE-dependent.

These results together suggest that brain amyloid levels may not result in neurodegeneration if microglia do not activate to their LDAM state. This requirement for LDAM status would explain why the PSEN1 mutation carrier with the Ch mutation had very high amyloid but limited tau and neurodegeneration. Future research on lipid droplet formation and AD pathogenesis includes (1) the characterization of the LDAM secretome and its effects on neurons in different genetic backgrounds, (2) screens to identify neurotoxic lipids being transferred from microglia to neurons, and (3) in vivo CRISPR screens to identify genes involved in lipid droplet handling in microglia in different APOE backgrounds.

Discussion

What lipid species are present surrounding amyloid plaques?

Performing single-cell lipidomics on human brain tissue is very difficult. Notably, several papers have shown that lipid droplets in in vitro and mouse models are triglyceride rich. However, the lipid droplets within cells may be filled with triglycerides, while secreted lipids are of different lipid species.

Could red oil O staining in human tissue also detect lysosomes? Amyloid only mouse models do not exhibit increased neutral lipid staining, but tauopathy mouse models to have increased neutral lipid staining co-localized with lysosomes.

Oil O could also be labeling lysosomes. A number of laboratories have shown in vitro that amyloid can induce droplets, but in vivo, a mix of dying neurons and myelin debris likely contribute to lysosomal lipid accumulation. Endogenous lipid synthesis could be occurring simultaneously in vivo.

In the co-culture experiment with conditioned media, is the neurotoxic effect the result of lysosome impairment or some other mechanism?

CRISPR screens are underway to identify genes necessary for this neurotoxic effect. Candidate genes are involved in uptake of lipids, trafficking to lysosomes, degradation of lipids in lysosomes, and transport of lipids to mitochondria. These candidate genes suggest there is a transport of lipids to neurons causing degradation.

What is the relationship between APOE expression level and lipid drop formation?

Some previous work shows that APOE localizes to lipid droplets and may regulate lipid droplet size. Dysfunctional metabolism of lipid droplet stores may be occurring in an APOE-dependent manner, but that potential mechanism would be separate from APOE-dependent trafficking and uptake.

Could the upregulation of ACSL1 be the result of lipid accumulation since ACSL1 is also mediating fatty acid oxidation?

This upregulation is likely due to an innate immune response given that LPS can also cause massive upregulation of ACSL1. Others have shown that macrophages in the periphery have a similar response to bacterial challenge where they upregulate ACSL1 and other similar enzymes to produce lipid droplets that have an antimicrobial host defense mechanism. This suggests that these microglia are mounting an innate immune response rather than a response to metabolism already occurring in cells.

Biology of APOE Protective Variants

Yadong Huang, M.D., Ph.D., Gladstone Institute/University of California, San Francisco

APOE3Ch and APOE2 variants confer protection against AD risk. To determine whether the Ch mutation protects against APOE4-related AD pathology, a mouse line was engineered to express human APOE4Ch at the endogenous mouse locus. Indeed, the Ch mutation appears to protect against tau pathology in PS19/E4 mice. However, two copies of APOE4Ch were required to reduce tau pathology to a level seen in PS19/E3 mice. The Ch mutation reduces APOE4-promoted tau uptake that is mediated by HSPG in human iPSC-derived neurons. In PS19 mice, APOE4Ch also protects against hippocampal atrophy in a gene dose-dependent manner and microgliosis with a dramatic effect of even a single APOE4Ch copy. Based on scRNA-seq data, APOE4Ch also protects against APOE4-induced microglial subtype changes in PS19 mice by increasing homeostatic microglia and reducing DAMs.

Human wildtype tau knock-in mice expressing APOE2 at the endogenous mouse locus (tau-KI/APOE2) have improved learning and memory performance compared to those expressing APOE3 (tau-KI/APOE3) as measured by the Morris Water Maze at 10 months of age. Memory was tested 72 hours after the last hidden trial. Deconstructing local field potential recordings into certain frequency components associated with cognition in the hippocampus revealed that tau-KI/APOE2 mice had higher CA1 sharp wave ripple-associated slow gamma power compared to tau-KI/APOE3 mice, a pattern predictive of better memory. Moving forward, additional in-depth studies are needed to characterize the underlying molecular and cellular mechanisms of the AD protective APOE variants.

Discussion

Is microgliosis altered in tau-KI/APOE2 mice?

Tau-KI/APOE2 mice exhibited a decrease in number of microglia at ten months of age compared to tau-KI/APOE3 mice. Analysis of microglia at 15 months and scRNA-seq data will provide additional information on microglia and microglia subclass abundance in tau-KI/APOE2 mice.

Are there differences in myelination between APOE4Ch and APOE2 variants?

At ten months of age, myelination immunostaining data indicate a trend toward higher levels of myelination in tau-KI/APOE2 mice compared to tau-KI/APOE3 mice. There may be significant myelination effects at older ages, and scRNA-seq of these models will uncover any oligodendrocyte changes between these variants.

Do APOE4Ch mice have fewer TERMs because of a lack of atrophy due to neurodegeneration?

The APOE4Ch model has decreased neurodegeneration and decreased microgliosis. The decrease of microgliosisde may contribute to the decrease in neurodegeneration.

Is APOE2 less protective than Ch in this model? Does APOE2 and Ch share a common protective mechanism?

These variants were tested in different models initially, so the level of protectiveness cannot be directly compared. Based on limited data, these two variants may have different protective mechanisms.

Would APOE2Ch have a stronger protective effect than APOE2 or Ch alone?

These two variants have not been combined yet, but they may have more protective effects together than in isolation.

Biological effects of APOE3ch on amyloid-induced tau seeding/spreading

Yun Chen, Washington University

A case report of APOE3Ch was the first instance of a variant protecting against autosomal dominant AD, and the individual’s onset of cognitive decline was delayed by 25 years. APOE3Ch heterozygotes can experience protective effects as well, albeit to a much lesser degree. While amyloid pathology in the APOE3Ch case was typical of autosomal dominant AD, tau pathology was absent, unlike non-APOE3Ch autosomal dominant AD cases. To better understand the protective effects of APOE3Ch, a humanized mouse line was created.

APOE3Ch mice have a variety of brain and peripheral phenotypes. In addition to hyperlipidemia, APOE3Ch mice have levels of VLDL not normally seen in BL6 mice. Notably, no lipid buildup was detected in CSF. In an amyloid model (APP/PSEN1), APOE3Ch exhibited reduced plaque formation, which was correlated with suppression of Aβ fibrillization in vitro. To assess the effects of this variant on tau seeding, human tau extraction was injected into the hippocampus and cortex, which recapitulates amyloid-induced tauopathy in humans with formation of neurofibrillary tangles (NFTs) and near-plaque tauopathy. APOE3Ch mice exhibited a dramatic reduction of Aβ-induced tau seeding, spreading, near-plaque tau pathology, and near-plaque tau-associated neuronal dystrophy. Together, these findings suggest that APOE3Ch blocks the formation of tau deposits and resultant neuronal damage.

Previous literature has shown that microglia activity around plaques is essential to rescue the near-plaque tau pathology in mice. Consistent with this previous observation, amyloid model APOE3Ch injected with human tau have increased microglia around plaques as well as increased lysosomal activity in those microglia. This enhanced microglial activity is not specific to tau injection—this phenomenon is also observed in mice without the injection and in brain regions where the tau injection did not spread. To further understand microglial changes due to APOE3Ch, bone marrow-derived macrophages (BMDMs) were isolated from mice to mimic microglia in in vitro functional studies. APOE3 and APOE3Ch BMDMs phagocytosed Aβ fibrils to a similar extent, while APOE3Ch BMDMs exhibited increased phagocytosis of human tau fibrils compared to APOE3 BMDMs. In addition, compared to APOE3 BMDMs, APOE3Ch BMDMs more readily degrade tau fibrils and release fewer tau seeds.

A competitive uptake assay was used to uncover why APOE3Ch BMDMs phagocytose more tau fibrils compared to APOE3 BMDMs. When APOE3 or APOE3Ch BMDMs were co-cultured with tau fibril and lipidated APOE3, tau uptake was suppressed. However, much less suppression occurred when co-culturing with lipidated APOE3Ch. Overall, these results suggest that tau and APOE may compete for the same receptors on cells, and tau can outcompete APOE3Ch.

Discussion

Is there a reduction of microglia in APOE3Ch mice?

The reduction in microgliosis is due to overall reduced pathology in APOE3Ch. Surrounding each plaque, microglia activity is enhanced in these mice.

How might APOE3Ch decrease tau usually stuck in dystrophic neurites due to a lack of axonal transport? How are APOE3Ch microglia reaching tau, or are these microglia taking up whole dystrophic neurites?

Based on current data, the possibility that microglia are phagocytosing neurites cannot be excluded. However, these mice exhibit no reduction of synaptic density around plaques. The protective effects of APOE3Ch are mainly observed in the tau injection model, and the current hypothesis is that microglia can phagocytose secreted tau seeds. However, experimental evidence currently cannot prove this hypothesis because monitoring tau seed secretion is difficult in in vivo models but possible in in vitro models.

The original Ch paper was in one person and used very atypical filtering that eliminated multiple potential protective variants. While the data in mice show a protective effect, there are currently no data indicating that Ch is protective in humans.

This conclusion might be challenged by the new publication of APOE3ch Hets (PMID: 38899694, DOI: 10.1056/NEJMoa2308583), despite the potential argument about statistic methods.

What are the astrocyte-related phenotypes in APOE3Ch mice?

Astrocytes near plaques in APOE3Ch mice do not exhibit unique phenotypes, although further work is needed to examine reactive astrocytes. Astrocytes express HSPG and LRP1, so there could be an analogous mechanism to reactive microglia. However, unlike DAMs upregulate proteoglycans such as SDC2, astrocytes do not upregulate HSPGs in response to amyloid pathology according to SEA-AD database, suggesting a different microenvironment in the presence of amyloid versus tau.

What are the expression levels of APOE in amyloid and tauopathy APOE3Ch mouse models?

No significant difference in APOE expression between APOE3 and APOE3Ch mice with and without amyloid pathology. In the periphery, APOE3Ch mice exhibit dyslipidemia with accumulation of APOE3Ch particles due to lack of LDLR-related clearance. Levels of APOE in APOE3Ch mice were not compared to APOE4 mice.

Is the lipid status in APOE3Ch mice similar to the hyperlipidemia observed in APOE knockout mice?

The comparability of lipid status depends on how hyperlipidemia is defined. If hyperlipidemia status is defined solely by cholesterol levels, then perhaps the lipid status in APOE3Ch and APOE knockout mice are comparable. However, the mechanism in APOE3Ch appears to be more similar to APOE2 mice because APOE knockout mice lack HDL-containing APOE seen in APOE3Ch and APOE2.

(Machine) Learning Features of the Alzheimer’s Disease Landscape

Olivier Lichtarge, M.D., Ph.D., Baylor College of Medicine

Dr. Lichtarge’s group computationally studies the unknown heritability and genetic backgrounds of various diseases to identify novel disease-associated genes, disease mechanisms, drug targets, and treatments. Recently, the group developed algorithms that connect evolution with fitness landscapes to identify differences in mutational energy between AD cases and healthy controls. These algorithms implement three principles: the work-energy principle, the distribution of energy following a Boltzmann distribution, and the equipartition theorem.

The work-energy principle states that energy (i.e., work) is equal to the distance a particle moves multiplied by the force applied to the particle. In terms of biology, distance refers to mutational change (e.g., movement from alanine to valine is small while movement from alanine to lysine is large) measured by substitution logarithmic odds. Force refers to the slope of a fitness landscape (e.g., mutations between nearby species are shallow, while mutations between distant clades are steep) measured by an evolutionary trace algorithm. These variables of biological force and distance can provide mutational energy, which is a proxy for the functional impact of a mutation and correlates well with experimental data on mutational impact. Next, Dr. Lichtarge’s group found that an ensemble of mutations follows a Boltzmann distribution, which is derived from statistical mechanics. Lastly, the equipartition theorem states that energy is distributed equally amongst all degrees of freedom in a physical system at equilibrium. By applying this theorem to genetics, in which any given gene has an equal amount of mutational energy, the group can compute the selection pressure (i.e., fitness effect) on a gene. Together, these three principles allowed the group to develop a genetics algorithm that can compute mutational energy and selection pressure following statistical mechanical patterns. Although other methods exist to compute mutational impacts, they do not follow a Boltzmann distribution and do not perform as well in calculating selection pressure compared to the group’s algorithm.

Dr. Lichtarge’s group used their algorithm to identify 122 novel genes associated with AD by analyzing AD and healthy control genetics data from the R2 release of the National Institute on Aging Genetics of AD Data Storage Site (NIAGADS). These novel genes cluster together in GWAS analysis, are associated with clinically catalogued and mouse model knockout AD phenotypes, and co-cluster with known GWAS-identified genes from relevant pathways, such as genes associated with amyloid, tau, lipid metabolism, immune response, and cell cycle. Comparisons of mutational energy of these genes revealed that some mutations in AD individuals have high mutational energy, indicating a pathogenic phenotype; conversely, some mutations in healthy individuals also have high mutational energy, indicating a protective phenotype. Furthermore, expression of these genes Drosophila models that overexpress either tau or Aβ42 either exacerbated or ameliorated the overexpression phenotype, suggesting a direct connection to phenotype pathology.

Dr. Lichtarge’s group also developed and trained a machine learning (ML) algorithm to distinguish between AD individuals and healthy controls using both known AD-associated genes and the novel genes identified from their mutational energy and selective pressure algorithm. The ML algorithm successfully predicted cases of AD and controls, but the prediction rate decreased when APOE3 and APOE4 training was removed. However, the ML algorithm more accurately predicted AD risk in individuals harboring APOE3 mutations compared to those harboring APOE4 mutations. Furthermore, AD risk genes in individuals harboring APOE3 tended to be pathogenic, while AD risk genes in individuals harboring APOE4 tended to be protective. In addition, the ML algorithm identified APOE3-, APOE4-, and sex-specific genes contributing to AD risk prediction. Together, these findings demonstrate the utility of the group’s ML algorithm in predicting AD risk with higher accuracy than existing ML models.

Discussion

Does ancestral diversity have an effect on the algorithm’s output, and how does this algorithm relate to polygenic risk scores?

The Lichtarge group is currently applying the algorithm to various ethnic groups from the NIAGADS R3 and R4 release cohorts, and the algorithm has preliminarily identified different AD risk genes between ethnic groups. The Lichtarge group is curious about which genes and implicated pathways overlap between ethnic groups.

This ML approach is different from applying polygenic risk scores but can complement both polygenic risk scores and GWAS data. In addition, the ML algorithm produces comparable results to polygenic risk scores. Future directions include using the ML algorithm to identify individuals who should be screened for diagnostics and interventions.

How is the force of amino acid movement in a mutation calculated, and does the algorithm account for amino acid position and gene length?

The algorithm normalizes the mutational energy calculations to gene length. The computational research field has repeatedly used this evolutionary trace algorithm to compute the evolutionary gradient. Furthermore, the evolutionary trace algorithm has been validated in a number of prior protein engineering studies to identify and compare functional sites.

Does the mutational energy calculation account for protein folding?

Variables such as protein folding, dynamics, compartmentalization, interactions, and ubiquitination are accounted for by evolutionary constraints and species divergence in these algorithms. The algorithms’ calculations average each of those variables across each species and the entirety of evolution, resulting in a first order approximation.

Have genes with different mutational energy between AD cases and controls been reproduced in independent datasets, and can the algorithm be applied to non-coding variants?

The Lichtarge group has tested the reproducibility of its mutational energy algorithm’s results, and the algorithm generally performs well compared to other methods, with few exceptions. The Lichtarge group is currently working on applying their algorithms to non-coding mutations.

Leveraging Deep Molecular Profiling to Understand APOE Dependent and Independent Pathology

Carlos Cruchaga, Ph.D., Washington University in St. Louis

For the past five years, Dr. Cruchaga’s group has conducted molecular profiling of human brain, CSF, plasma, and iPSC samples. The group aims to use proteomics, metabolomics, transcriptomics, lipidomics and other omics data to identify new biological processes associated with disease, create new prediction models, and identify new therapeutic targets. The group obtained over 3,000 samples from individuals with sporadic AD, Mendelian AD, and risk variant carriers, and implemented a proteogenomic approach to their analysis to further elucidate APOE’s involvement in AD.

Dr. Cruchaga’s group conducted proteomic assays using the SomaLogic platform, which analyzes over 7,000 proteins from CSF samples, and found that APOE protein levels are higher in the CSF of individuals with early-stage AD (i.e., amyloid positive and tau negative) compared to those with late-stage AD (i.e., amyloid and tau positive). These APOE protein levels are independent of APOE genotype, indicating a potential for researchers to identify biomarker signals that are more dependent on AD stage pathology than APOE genotype. In addition, the group conducted a regulatory genetic region analysis, which compared protein position to protein quantitative trait locus. This analysis revealed that the APOE locus is highly pleiotropic. Furthermore, over 400 proteins in that region are associated with APOE2 and APOE3 expression, enriched in neurons compared to glia cells, and associated with apoptosis and long-term potentiation pathways. Analysis of cognitively normal and amyloid- and tau-negative individuals revealed similar elevated expression in these proteins, indicating that APOE is interacting with neurons before AD pathology is present. These newly identified proteins can potentially be used for research on target engagement and biomarker outcomes in clinical trials in the future.

Discussion

How early are significant changes observed in APOE4 levels, and can this group’s approach be beneficial for developing hypotheses about timing?

Samples used in this work represent a broad age range of individuals (i.e., 45 to over 90 years old), including individuals of varying AD disease stage. However, the group’s work has not yet assessed at which specific timepoints these protein levels begin to change. Based on analysis of cognitively normal individuals, Dr. Cruchaga hypothesizes that these changes occur in the early stages of AD.

Are there differences in APOE2, APOE3, and APOE4 protein levels in CSF samples, and is there a main effect of genotype on CSF protein levels in healthy controls?

The protein level differences across various APOE genotypes are complicated and dynamic. Initially, APOE protein levels increase with disease progression, but then decrease in late-stage AD. However, this analysis was not conducted in healthy control samples.

Is there a potential association of APOE protein levels with NfL protein levels?

The Cruchaga laboratory published a paper six months ago that discusses comparisons of APOE and NfL protein levels.

Which has a larger impact on AD pathology, APOE protein levels or APOE genotype? Have strong regulators of APOE protein levels in CSF been identified in other ancestries?

The original regulatory genetic region analysis to determine APOE pleiotropy did not correct for APOE itself. The original analysis was conducted with CSF samples from European individuals, but the Cruchaga laboratory has 200 CSF samples from African American individuals for future analyses.

What was the underlying explanation for APOE levels decreasing in later stages of AD?

Other biomarkers such as pTau also decreased in later stages of AD, but the specific mechanisms for this change are not clear. These changes may be related to neuronal death in later stages of AD.

Are SomaScan or APOE slow off-rate modified aptamers (SOMAmers) robust and accurate enough to confidently rely on the observed CSF findings?

The SomaLogic platform provides several SOMAmers for APOE, including one targeting general APOE and others targeting specific APOE proteoforms. Observing a C signal for protein levels provides evidence that aptamers are specifically detecting APOE proteoforms. However, the genotype-specific SOMAmers may not provide sufficient detection specificity because, for example, APOE4 SOMAmers often measure general APOE and APOE2. Not understanding which proteoforms are being measured by analysis platforms is a current challenge in studying APOE protein levels.

Post meeting comment from Carlos Cruchaga: SomaLogic has several APOE assays. Some of them are labeled as APOE-2, APOE3 and APOE4 assays. I agree with that comment that those assays are in fact not-proteoform specific. I do not think the assay for APOE2 is only measuring APOE-2 levels and so on. I think however that all the APOE assays are measuring APOE levels in general. This is based the observation that when we performed QTL mapping, we found a strong signal in cis. Specifically, the variants that codify for APOE2 and 4, as the strongest signals for APOE-assays.

So, we think we are measuring APOE in general, but not the specific proteoforms.

APOE4 Impact on Vasculature

Sally Temple, Ph.D., Neural Stem Cell Institute

Cerebrovascular pathology is an early and common factor in AD and related dementias and is exacerbated by the APOE4 genotype. APOE4 carriers treated with amyloid-reducing medication also have an increased risk of ARIA such as edema and hemorrhage. APOE4 may contribute to ARIA through changes in baseline membrane metabolism and inflammation and amyloid accumulation in vessels resulting in cerebral amyloid angiopathy (CAA). Prior studies have also found that APOE4 carriers who are cognitively normal exhibited BBB breakdown according to MRI, which was exacerbated by disease onset and resulted in peripheral damage. Furthermore, both aged APOE4 and APOE4;5XFAD mouse models showed BBB breakdown, reduced cerebral blood flow, neurodegeneration, and behavioral deficits. In addition, prior immunohistochemistry (IHC) studies have shown that APOE expression is strong in vascular cells in both the mouse and human brain, notably in CAA.

Dr. Temple’s group, led by Dr. Taylor Bertucci and their collaborators Drs. Oscar Harari, Ricardo D’Oliviera Albinus, and Celeste Karch, aims to understand the impact of APOE4 on gene expression patterns in cerebrovascular cells in the brain. This has been historically challenging to study by single cell transcriptomics due to the rarity in capturing brain endothelial cells . Drs. D’Oliviera Albinus and Harari analyzed data from several studies, totaling over 130,000 cerebrovascular cell nuclei. This large-scale analysis enabled them to identify endothelial cells (ECs), pericytes, fibroblasts, and smooth muscle cells (SMCs), all of which expressed APOE. The transcriptome analysis revealed that ECs from healthy APOE4 carriers showed increased interferon gamma (IFNG) pathway signaling and decreased oxidative phosphorylation compared to healthy APOE3 carriers. Further analysis revealed that guanylate binding protein (GBP) 2 and GBP4, whose activity is induced by interferons, were upregulated and enriched in vascular cells. In addition, genes that have known associations with AD were upregulated in vascular cells from APOE4 carriers compared to APOE3 carriers.

Dr. Temple’s group then collected a set of isogenic induced pluripotent stem cell (iPSC) lines with APOE3/3 vs APOE4/4 genotype through collaboration with Drs. Julia TCW, Alison Goate, Jeffery Vance and Derek Dykxhoorn. Using a recently published protocol PMID: 34591294, Dr. Temple’s group developed isogenic iPSC-derived vascular cells (ECs and various mural populations including pericytes and smooth muscle cells) with identical genetic backgrounds harboring APOE4/E4 (i.e., homozygotes) and APOE3/E3. In a prior study led by Drs. Joel Blanchard and Li-Huei Tsai, APOE4/E4 pericyte-like cells were found to exhibit increased nuclear factor of activated T cells (NFAT) and calcineurin signaling compared to APOE3/E3 cells. Blanchard et al PMID: 32514169.

Following this exciting finding in pericytes, Dr. Temple’s group focused on APOE3/3 vs APOE4/4 ECs. Before an inflammatory challenge, the group found differences between APOE4/E4 and APOE3/E3 ECs across various pathways, including molecules that are commonly increased in the brains of APOE4 carriers. After an inflammatory challenge, APOE4/E4 ECs showed elevated IFNG and pro-inflammatory gene profiles, elevated endothelial-mesenchymal transition, decreased protocadherins implicated in barrier function, and altered transporter expression. Using these ECs, the group also developed two-dimensional monolayer cultures to model and assay barrier function. After an inflammatory challenge, Drs. Kate Tubbesing and Liz Fisher observed that barrier resistance was significantly decreased in APOE4/E4 EC monolayers compared to APOE3/E3 ECs monolayers, and by IHC analysis, significant endothelial barrier breakdown was observed in APOE4/E4 but not in APOE3/E3 EC monolayers. Furthermore, APOE4/E4 EC monolayers expressed higher extracellular matrix-related genes compared to APOE3/E3 EC monolayers, including FN1. The group is currently developing novel three-dimensional vascular plexus models that include venous- and arterial-like vessels, pericytes, and SMCs for further research on APOE4 impact on the vasculature. PMID: 37564277.

Discussion

Does FN1 upregulation appear only in ECs, or does it also appear when ECs are co-cultured with pericytes?

Those FN1-upregulation data were from purified ECs, but future work will explore this in mural cells. After revisiting our data, we also found that collagen type VI alpha 3 chain was elevated, and heparin binding EGF-like growth factor was reduced. This corroborates findings made in brain tissue by Drs. Caghan Kizil and Richard Mayeux reported at the workshop, which has led to a new collaboration.

How were type 1 and type 2 interferons distinguished in this project? Also, is there clinical evidence that APOE4 carriers show differences compared to APOE3 carriers when they face a viral infection that can penetrate the brain?

Dr. Temple is planning to explore this inflammatory response and various factors involved. Whether these inflammatory differences between APOE4 and APOE3 carriers are also observed in response to viral infections remains unclear. Some literature may suggest an association between parasitic infection and protection conferred by APOE4.

Has the Temple group explored transcriptomic differences between APOE3 and APOE4 in response to amyloid beta?

Exploration of response to Aβ is ongoing.

Has the Temple group conducted lipid droplet analyses in their two- and three-dimensional models? Also, did these analyses include an APOE4 knockout iPSC line?

IHC staining for lipid droplets is ongoing, and an APOE4 knockout iPSC line was not included in the presented analyses.

Do these inflammatory data suggest that BBB breakdown in APOE4 carriers measured by MRI is primarily triggered by early inflammatory changes? How would APOE4 cause such early changes?

BBB breakdown is potentially indicative of an early inflammatory state that contributes to the reduction of the BBB, which the Temple group and collaborators are exploring.

Why are pericytes influenced by APOE, and what is the impact on BBB function?

Joel Blanchard’s (Icahn School of Medicine) recent paper demonstrates changes in the NFAT-calcineurin pathway in pericytes, and prior studies have shown how pericytes are critical for stabilizing vasculature. The Temple group has also identified several differentially expressed genes in mural cells, so researchers will likely be interested in exploring those pathways and how those cells alter endothelial barrier function. The three-dimensional models currently in development will enable future exploration of these cell type interactions.

ApoeE2 and Its Role in Plaque Deposition, Neuroinflammation, and Neurodegeneration

Bradley Hyman, M.D., Ph.D., Massachusetts General Hospital

APOE4 is strongest known risk factor for AD and associated with an increase in plaque numbers, synaptic loss, and oligomeric Aβ. Furthermore, APOE4 impacts tau-mediated neurodegeneration through microglial-mediated mechanisms. APOE genotype also impacts microglial set point and the clinical rate of AD progression, even after accounting for genotype effects on plaques and NFTs. However, recent studies provide evidence that the APOE2 allele is protective against AD progression. Prior large population studies have shown that inheritance of a single allele of APOE2 was protective as evidenced by increased age of AD onset by approximately 20 years. In addition, Dr. Hyman’s group’s analysis of several clinical databases indicated that APOE2 is associated with milder clinical and neuropathological AD phenotypes, including decreased plaques and NFTs and slower AD progression. Recent human transcriptome and single nucleus (sn)RNA-seq data suggest that the impact of APOE4, and the relative protection from APOE2, may be due to effects on microglial activation.

Using APOE2 as a potential therapeutic for AD is currently challenging because APOE is a soluble protein in blood and CSF but does not cross the BBB; APOE is produced in the brain by CNS cells and not in the periphery; and the entire brain requires APOE2 treatment because APOE is expressed in all brain regions. However, a possible solution is injecting APOE2 genetic material into the ventricles, allowing transduction of the ependyma and potential secretion into the parenchyma and CSF, thus bathing the brain in gene product.

Dr. Hyman’s group administered AAVs carrying APOE2 mRNA at three escalating doses into mouse brains, finding that all doses transduced the ependyma and only 10 percent of endogenous APOE secreted into the parenchyma. When APOE2 AAVs were administered into the brains of mice endogenously expressing APOE4, the group observed a dose-dependent decrease in plaque deposition, as measured by IHC 60 days after injection. Furthermore, enzyme-linked immunosorbent assays (ELISAs) showed a similar dose-dependent decrease of Aβ42 protein levels. APOE2 administration into APOE4 mouse brains also prevented microglial activation near remaining plaques in a dose-dependent manner, as measured by microglial morphology and biomarkers. Lastly, the APOE4 mouse brains showed amelioration of synapse loss near plaques, as measured by a decrease in oligomeric Aβ42 levels, and nearly complete restoration of synapses compared to vehicle controls. Together, these findings suggest that APOE2 protein acts in a non-cell autonomous manner and mimicking APOE2 in the CNS may be therapeutically beneficial.

Discussion

Because Dr. Hyman was not present at the workshop in person, all questions (listed below) were sent to Dr. Hyman at the conclusion of the workshop:

My question is when you overexpress APOE2 in the brain have you looked at what level the endogenous APOE4 protection is being suppressed, is it suppressed at all and to what extent?

The mass spec data do not show a statistically significant reduction in APOE4, but we would have been underpowered to see a small effect.

My question is that if you look at genetic data right now from patients, it's a very rare indication where APOE3 now in a context of APOE4. It seems to be this is doing this progression. We have a lot of data demonstrating that knock out APOE3 positively impacted amyloid plaque formation. What do you think is going on with your data where you show that overexpression of APOE2 positively impacted the biology?

My working hypothesis is that apoE has two roles- one is to stabilize amyloid fibrils (it does this for Abeta as well as numerous other classical amyloid proteins), and a unique role in setting microglial inflammatory tone. We are trying to figure out how it does that, but as a first blush it is in a cell nonautonomous fashion, and our human data are consistent with the idea that E4 leads to increased neuroinflammation and E2 to less given a CNS insult.

What is the amount of APOE2 expressed by endogenous levels of APOE4 and additionally are the APOE2 lipidated. Do you have plans to look at the Christchurch variants on effect of E2, E3, E4 and question: We expect AD patients to have some level of disrupted blood brain barrier, would peripheral injection be helpful in these cases to reach the brain and improve AD outcomes? Any experimental data available from your group?

We have looked in mice and peripherally injected apoE does not seem to cross the BBB in mouse amyloid models in detectable amounts.

The E2 is ~10% ish of endogenous levels, which suggests to me that it is not simply competing with E4 but in fact doing something positive.

We don’t have any data on Christchurch etc.

Clinical Trial of APOE2 Gene Therapy

Ronald Crystal, M.D., Weill Cornell Medical College

Addressing the underlying genetics of the AD brain by converting the APOE4 homozygote brain to an APOE2/E4 brain may correct many of the downstream biological mechanisms underpinning the disease. Dr. Crystal’s group developed a clinical trial strategy involving the administration of an AAVrh.10 serotype coding for the human APOE2 gene (LX1001 [i.e., AAVrh.10hAPOE2]), which contained a cytomegalovirus early enhancer/chicken β actin promoter that functions in all cell types across the human body. For preclinical testing of LX1001, all mouse models the group used in experiments were developed on a Transferrin receptor-like protein 4 (TRE4) genetic background. Initially treating APP/PSEN1 mice expressing human APOE4 with an intra-hippocampus delivery of LX1001 decreased insoluble Aβ42 protein levels in the brain in a dose-dependent manner 4.5 months after injection. However, the Crystal group needed to explore intracisternal administration methods of LX1001 delivery as a more feasible route for human patients. Intracisternal administration of LX1001 in non-human primates (NHPs) evenly distributed APOE2 vector, mRNA, and protein throughout the brain.

Dr. Crystal’s group’s Phase 1 clinical trial of LX1001 treatment recruited a study population of APOE4 homozygote patients over 50 years old with disease stage ranging from mild cognitive impairment to moderate dementia, as confirmed with CSF biomarkers and amyloid positron emission tomography (PET) scans consistent with AD. The trial’s primary endpoint was safety profiling, with secondary endpoints assessing conversion of CSF from APOE4/E4 to APOE2/E4, amyloid PET scans, quantitative MRI, cognitive testing, and measurement of CSF Aβ42, total Tau (tTau), and pTau protein levels, all of which were assessed at 12-months after AAV administration. Trial participants received three intracisternal LX1001 doses based on patient CSF volume quantified by MRI. No immunosuppressives were administered in the first cohort, but subsequent cohorts received prednisone for eight weeks. The administration procedure produced no serious adverse events, but transient headache related to the administration procedure occurred in two of the five participants.

According to preliminary analysis of 12-month follow-up data, although mass spectrometry analysis showed that APOE4 protein levels remained at relatively endogenous levels, Cohort 1 participants continued to express APOE2 after AAV administration. In addition, CSF biomarker data showed decreases in Aβ42 protein levels in 2 out of 3 participants and decreases in tTau and pTau protein levels in all 3 participants who participated in 12-month follow-ups. The group is currently analyzing data from other LX1001 dosing groups. The clinical trial is now being run by LEXEO Therapeutics, and LEXEO presented findings during the Clinical Trials on AD conference in October.

The group has also assessed other approaches to APOE gene therapy in additional preclinical studies. First, compared to APOE2 alone, APP/PSEN1 mice treated with an AAV carrying the APOE2Ch variant showed suppressed CNS amyloid pathology, and P301S mice showed suppressed CNS tau pathology. Next, AAV-directed anti-APOE micro (mi) RNA CNS expression suppressed CNS APOE4 protein levels in mice expressing human APOE4 and TRE4. Lastly, AAV-directed prime editing mediated the conversion of APOE4 to APOE3 in mice expressing human APOE4.

Discussion

According to the protein CSF measurements after AAV administration in the clinical trial, less than one percent was APOE2 protein. What is the target APOE2 protein CSF level?

Achieving higher APOE2 protein levels in the mouse brain is easier than in the human brain because mouse brains are smaller. APOE2 protein levels in humans are expected to reflect the dose dependency observed in NHPs.

Given that APOE4 may be beneficial for age-related macular degeneration and glaucoma, did the Food and Drug Administration (FDA) request any vision testing on trial participants?

The FDA did not request any vision testing. Separately, Dr. Crystal has corneal nerve data in AD patients unrelated to gene therapy. When these AAVs are administered to the CNS, PET scans showed that 50 to 60 percent of vectors leave the CSF, indicating systemic spread. However, when these AAVs are administered intravenously (IV), it does not spread to the eye.

Given that these data suggest AAV leaking from the CSF to the liver, are there any peripheral impacts from AAV administration?

No hypolipidemia has been observed in these studies. When these AAVs are intravenously administered, 60 to 70 percent of the vector spreads to the liver. However, no systemic effects in the clinical trial have been reported thus far.

Do other methods exist to adequately administer the AAVs other than intracisternal administration?

In general, three different approaches are used for AAV administration: (1) a direct catheter to the brain, (2) within the ventricle to the CSF, and (3) to the cisterna magna between C1 and C2, which is the approach used in this clinical trial. The field has shown interest in systemic administration and developing vectors to cross the BBB. In NHP studies of systemic administration, the vectors showed adequate coverage throughout the brain, but that coverage required high intravenous doses.

What efficacy does the clinical trial achieve with APOE4 to APOE3 conversion? Also, what information does your group communicate to patients about receiving gene therapy?

Preclinical mouse data showed an approximately 20 percent conversion of APOE4 to APOE3. While the APOE2/E4 genotype is protective against AD progression, the level of APOE2 protein needed for this protection remains unknown. The advantage of gene therapy is that it only requires one-time administration and results in long-term and local expression.

In which cell types do these AAVs transduce, and has APOE2 integration been observed?

In NHPs, these AAVs tend to transduce more neurons than glia, although the Crystal group has modified the vector to transduce more glia. Some APOE2 integration and liver malignancies were observed in mouse models, but integration has never been observed in humans.

Is an AAV caudal injection and placing then the patient in a decline position feasible?

Based on previous experience, administering the AAV lower in the spine is not as effective as administration between C1 and C2.

Given that APOE2 and the Ch variant may offer different mechanisms of AD protection, is there any basis to co-inject APOE2 and Ch?

The Crystal group has cloned APOE2Ch into the same expression cassette. Some prior studies describe administering APOE2 and Ch in two separate vectors simultaneously.

Do the individual biomarker reductions of CSF Aβ42, tTau, and pTau correlate with the individual expression of APOE protein in the CSF?

Preliminary observations suggest a correlation between biomarker reduction and APOE protein expression, but the dataset lacks sufficient power for this correlation analysis.

Is the final presentation slide suggesting that AAV injection of the APOE3Ch variant is more effective than APOE2?

In mouse models, the administering the APOE2Ch variant is more effective in tauopathy mouse models compared to APOE2, and this observed efficacy is greater in tauopathy mouse models than amyloidosis mouse models.

What is the immune response of patients injected with these AAVs?

Researchers have shown interest in using immunosuppression prior to AAV administration. In a prior study on Batten disease in children, neutralizing antibodies against the vector were detected systemic circulation but not in the CSF likely due to the BBB. However, researchers are beginning to use immunosuppression for 8 to 14 weeks prior to AAV administration, although immunosuppression efficacy in the CNS remains poorly understood.

Potential Therapeutic Role for Peripheral APOE

Guojun Bu, Ph.D., Hong Kong University of Science and Technology

Various APOE variants, such as the Christchurch and Jacksonville variants, reduce AD risk by limiting APOE binding to heparan sulfate, inhibiting tau pathological propagation, reducing APOE self-aggregation, and enhancing lipid efflux. These protective functions can counteract APOE4’s gain-of-toxic function effects which include amyloid seeding, increased BBB leakage, decreased vascular function, and impaired insulin signaling, glucose metabolism, and mitochondrial function. APOE4 also exhibits loss-of-physiological functions, including decreased lipid metabolism, compromised microglial responses, impaired synaptic function, and reduced capacity in injury repair.

APOE protein is expressed in multiple cell types in the brain and periphery. In the CNS, APOE is expressed primarily in astrocytes and activated microglia; in these glial cells, APOE associates with HDL-like particles, transports lipids to neurons, and undergoes minimal de-sialylation. In the periphery, APOE is expressed primarily in the liver and macrophages, is detected at ten times higher concentrations in plasma compared to CSF, undergoes extensive de-sialylation, and associates with beta-very-low-density lipoprotein (βVLDL) and a subclass of HDL. Furthermore, APOE4 is a risk factor for hypercholesterolemia and atherosclerosis in peripheral tissues. Prior epidemiological studies have found that low plasma levels of APOE are associated with increased risk of AD and all types of dementia. In APOE3/E4 heterozygous individuals, APOE3 protein accounted for the majority of APOE isoforms detected in plasma. However, individuals with higher plasma APOE4 to APOE3 ratios had reduced grey matter volume and glucose metabolism.

To functionally explore the effects of peripheral apoE isoforms, the Bu group developed a mouse model where apoE expression is driven by liver hepatocyte-specific albumin Cre driver in an Apoe knockout background, thus allowing expression of human APOE3 or APOE4 specifically in liver hepatocytes alone. Using this model, the Bu group found that APOE4 expressed in liver hepatocytes resulted in compromised BBB integrity, reduced cerebral blood flow, and impaired cognitive and synaptic function compared to the Cre-negative littermate control mice which are equivalent to Apoe knockout. In contrast, APOE3 expression in liver hepatocytes in mice showed enhanced memory performance in a fear conditioning assay, as well as enhanced synaptic plasticity compared to their littermate controls. Vascular-enriched single-cell (sc)RNA-seq analysis revealed reduced astrocytic endfeet and increased gliosis associated with albumin Cre-driven APOE4 mice. In addition, analysis of the plasma protein network revealed that liver-specific expression of APOE3 increased expression of several protease inhibitors, such as tissue inhibitor of metalloproteinase 3 (TIMP3) and serpin protease inhibitor family E member 2 (SERPINE2). In contrast, liver-specific expression of APOE4 decreased expression of proteins involved in lipid metabolism and increased expression of proteins involved in inflammation. Lastly, the group bred these albumin Cre-driven APOE4 and APOE3 mice with amyloid-expressing mice, and found that mice expressing APOE3 had reduced amyloid plaque deposition while mice expressing APOE4 had increased amyloid plaque deposition. Together, these findings suggest that peripheral expression of APOE4 impairs brain function by compromising cerebrovascular integrity and function and by promoting inflammation.

Discussion

Have peripheral macrophages been characterized in the albumin Cre-driven APOE4 mice, and do these macrophages contribute vascular inflammation?

Although these peripheral macrophages have not yet been characterized, macrophages may potentially contribute to the proteomic profiling results. Macrophages also express APOE, so researchers can express APOE in macrophages alone to study their contribution to AD pathology and pathways.

Given that liver-specific APOE4 is disrupting the BBB, is APOE4 potentially spreading to the brain itself in these mouse models?

No, in our mouse model APOE protein was not detected in the brain by IHC, ELISA, and western blot assays. Mice younger than six months of age did not exhibit BBB disruption, indicating that vascular degeneration is occurring in older mice.

Could plasma APOE protein cross the BBB in the presence of disease?

The existence of non-lipidated APOE is uncommon, and lipoprotein particles bound to APOE are large, so APOE crossing the BBB is unlikely. Under disease conditions, APOE protein could potentially spread to the brain, but is unlikely to have a significant impact on brain function or pathology.

Is there evidence of immune cell infiltration into the brains of albumin-Cre-driver APOE4 mice?

Immune cell infiltration has not been explored.

Could transplantation of a liver expressing APOE3 into APOE4-positive patients improve brain or cognitive outcomes after surgery?

Several researchers are exploring APOE3 liver transplantation in APOE4 carriers. An early inspiration for the research presented was a transplant patient carrying APOE4 who received an APOE3-expressing liver. Significant evidence showed that APOE can impact vasculature and indirectly impact brain function.

Is targeting the CNS and liver together for APOE gene therapy the best approach as opposed to targeting the CNS alone?

Targeting the CNS and liver together is a beneficial approach because APOE4 is also a risk for atherosclerosis. However, APOE2 is associated with vascular issues, so researchers should be cautious about administering peripheral APOE2 gene therapies.

Given that these mice did not express APOE in the brain, what would occur if APOE isoforms were overexpressed in mice that express APOE at normal levels?

The Bu group is currently conducting experiments in mice overexpressing APOE isoforms in the presence of endogenous APOE. In addition, after injection of APOE3 or APOE4 plasma from young mice into older wildtype mice, the APOE3 plasma enhanced BBB integrity, while APOE4 plasma disrupted BBB integrity.

Antisense Oligonucleotides for Alzheimer’s Disease—A Focus on APOE

Hien Zhao, Ph.D., Ionis Pharmaceuticals

Gene therapies can use ASOs to target mRNA to either increase target gene expression or degrade target gene mRNA to reduce gene expression. To reduce gene expression, ASOs generally employ one of two mRNA degradation approaches: (1) ASOs will bind the target mRNA within the nucleus and recruit RNase H to cleave the target mRNA, or (2) ASOs will bind the argonaute 2 (Ago2)-Dicer protein complex and subsequently bind and cleave the target mRNA in the cytoplasm. ASO technology has evolved due to recent advances in medicinal chemistry, such as ASO backbone and sugar modifications that decrease pro-inflammatory sequelae, enhance ASO targeting to specific cell types, and increase binding affinity to target mRNA and stability against nucleases. For example, novel backbone mesyl phosphoramidate provides increased molecular stability to enable longer-term efficacy per dose.

Because ASOs do not cross the BBB, they are typically administered through lumbar cistern intrathecal injection. Following injection, ASOs distribute rapidly throughout the CSF along the neuroaxis, which can be enhanced with convection. ASOs then associate with the meningeal layer covering the CNS and penetrate cerebral arteries. Subsequent ASO movement into and within the parenchyma likely involves perivascular pathways, white matter tracts, and direct migration through the glial limitans via gap junctions or transcellular exchange mechanisms. Once distributed to various CNS regions, ASOs are endocytosed into cells to perform their mRNA-targeted action. The CNS can then clear ASOs through traditional CSF clearance pathways including meningeal lymphatics, peripheral lymph nodes, and then systemic circulation.

Dr. Zhao’s group has conducted several snRNA-seq studies to confirm ASO activity across cell types in the CNS. For example, administration of an ASO against metastasis-associated lung adenocarcinoma transcription 1 (MALAT1) mRNA in NHPs reflected mRNA knockdown at the single cell level compared to vehicle controls, although to various extents in different CNS cell types. Furthermore, the group found that administering an ASO against prion protein (PRNP) mRNA resulted in PRNP knockdown in the brain and CSF, indicating that CSF protein biomarkers can provide sufficient evidence of target protein knockdown.

Recently, Dr. Zhao’s group conducted studies to assess the impact of anti-APOE ASOs on mouse model brain pathology. When injecting AAP/PSEN1 APOE4-expressing mice with an anti-APOE ASO, the group found that a 50 percent suppression of APOE mRNA in neonates results in reduced plaques and dystrophic neurites by 16 weeks of age compared to vehicle controls. However, when the group administered the ASO in 6-week-old mice, plaque burden was not significantly reduced compared to vehicle controls. In another study using TE4 mice, the group found that injecting their anti-APOE ASO in 6-week-old mice reduced tau pathology, neurodegeneration, and plasma NfL (i.e., a marker of neuronal injury) and increased synapse numbers after 3 weeks compared to vehicle controls. Furthermore, the anti-APOE ASO decreased neuroinflammation as measured by reduced synapses engulfed by microglia and reduced astrocytosis, microgliosis, and cytokine levels. Together, these findings indicate that lowering APOE levels with anti-APOE ASOs ameliorates key pathologies and neuroinflammation in models of AD.

Discussion

When will this ASO enter clinical trials?

Hopefully this ASO will enter clinical trials in the future.

Endogenous APOE was knocked down in these mouse models, correct? Why did the ASO reduce the number of dendritic neurites but not the plaque deposition?

Dr. Holtzman explained that amyloid is likely toxic to neurites and results in dystrophy, and APOE attachment to amyloid may add another level of toxicity. Prior studies in humans have shown that more synaptic loss occurs around amyloid plaques in APOE4 carriers. The ASO may have resulted in less APOE attached to plaques, thus reducing the number of dystrophic neurites but not impacting the number of plaques.

Did the anti-APOE ASO reduce APOE more in astrocytes or microglia, and can those findings inform the targeting of specific cell types?

This ASO reduced APOE in astrocytes by approximately 50 percent and reduced APOE in microglia by approximately 60 percent. The ASO targets every cell type but targets some cells types better than others. Due to the inability to quantify ASO uptake by cell type, cell type targeting remains poorly understood.

Have other methods besides lysosomal marker staining also indicated a change in the number of dystrophic neurites?

Dr. Holtzman’s group has measured the number of dystrophic neurites after ASO administration by staining other molecules that accumulate in dystrophic neurites and found similar results. Thus, the ASO is likely reducing the number of dystrophic neurites and not just the number of trafficked lysosomes.

Do cell type-specific ASOs cause different effects on AD pathology, and can ASOs target specific isoforms of APOE?

Currently, this ASO targets all cell types, so cell type-specific targeting is dependent on the ability for each cell type to take up the ASO and normal gene target expression within that cell type. Targeting specific APOE isoforms with ASOs is difficult because each isoform only differs by a small number of amino acids.

Would greater than 50 percent APOE knockdown in this mouse model provide more benefits, and has Ionis Pharmaceuticals assessed APOE knockdown in the liver?

Dr. Khvorova’s presentation highlights other approaches to reducing APOE by more than 50 percent. Because APOE exhibits specific functions in the liver, Ionis was cautious about knocking down APOE too much. Moreover, no liver cholesterol level changes were observed in these models.

Has Ionis assessed APOE protein knockdown these mouse models, and does that knockdown correlate with mRNA suppression?

APOE mRNA and protein have approximately a one-to-one correlation; this ASO knocks down approximately 50 percent of APOE protein.

RNAi Modulation of ApoE: Delicate Balance between Plaque Clearance and Glia Activation

Anastasia Khvorova, Ph.D., University of Massachusetts

Currently, six small interfering (si)RNA-based drugs are approved that implement chemically modified siRNA conjugations. These approved siRNAs utilize either N-acetylgalactosamine conjugates that are recognized by glycoside receptors of liver hepatocytes, lipophilic siRNAs that are administered systemically, or antibody peptide conjugates. These siRNAs require chemical stabilization with a combination of sugar and phosphate backbone modifications to ensure a 6- to 12-month clinical efficacy from a single intrathecal injection. To silence APOE expression, Dr. Khvorova’s group developed a divalent siRNA (di-siRNA) that was broadly distributed throughout mouse and NHP brains after a single intracerebroventricular injection. Furthermore, the di-siRNA was broadly distributed throughout NHP brains after CSF delivery, which resulted in 80 to 90 percent silencing of APOE expression, and the level of di-siRNA accumulation was correlated with the duration of protein silencing.

Dr. Khvorova’s group initially assessed the silencing effects of their di-siRNA in APP/PSEN1 and 5xFAD mice after intracerebroventricular injection, finding that 80 to 90 percent of APOE mRNA and protein were silenced. Furthermore, APOE silencing continued 2 and 4 months post-injection, reduced Aβ42 and insoluble tau burden, and reduced lysosomal APP despite not reducing intracellular APP expression. Although the group detected reductions in liver APOE mRNA and protein and high dose levels, they did not detect reductions in circulating cholesterol levels. Even still, to address any potential impact on the liver, the group also implemented a combinatorial siRNA approach to selectively silence APOE in the brain alone, resulting in no systemic impact from APOE silencing. Modulating the dose for the CSF administration alone is likely to be sufficient to achieve CNS-selective ApoE modulation.

Prior to di-siRNA injection in APP/PSEN1 mice, APOE and Aβ42 co-aggregated as evidenced by brain tissue staining, which likely protects plaques from microglial degradation. After di-siRNA injection, the group found that APOE silencing significantly decreased the number of activated astrocytes. Microglia cluster around plaques and silencing of Apoe enabled activation of microglia and subsequent glia-based plaque clearance. Furthermore, RNA-seq analysis detected activated microglial signatures that were correlated to amyloid plaque burden. The observable activation of microglia signature was more correlated with healthy microglia function but raised concerns that excessive inflammatory responses in the brain might be harmful. The Janus kinase 1 (JAK1) is a major regulator of interferon response signaling. Thus, the group developed a dual-targeting siRNA scaffold that targets both APOE and JAK1 to simultaneously suppress APOE expression and reduce inflammatory responses in the brain. The group found that silencing JAK1 ablated a significant part of APOE silencing-induced glia hyperactivation in the CNS of wildtype mice while maintained significant levels of amyloid pathology clearance, indicating that hyperactivation of microglia is not necessary for productive amyloid clearance.

Discussion

How does silencing JAK1 signaling impact the inflammatory response?

Interferon signaling is potentially occurring as a secondary microglial function rather than a primary function. When microglia begin clearing molecules, they send an overexaggerated immune response signal. However, if siRNAs can stop that secondary immune response function of microglia, then plaque clearance can still occur.

When the di-siRNA silences APOE, does the microglial response damage synapses?

Dr. Khvorova’s group has not observed synaptic damage, although her group does not have sufficient expertise with these assays. The lack of synaptic damage could be because damage from plaques is significantly more severe than damage from microglia activation. Dr. Khvorova would need to collaborate with the Holtzman group to properly answer this question.

Is the act of removing plaques by microglia stimulating the interferon pathway, or could loss of APOE expression in microglia contribute to the interferon response?

The Khvorova group conducted an experiment in which APOE was silenced in wildtype mice and detected an interferon response. However, these findings contradict the findings from Dr. Holtzman’s group because APOE knockout mice did not show an increased interferon response. These conflicting findings may be due to timing, in which silencing APOE in wildtype mice may induce an interferon response compared to mice that have never expressed APOE.

Did the Khvorova group measure the levels of freeform or fused Aβ as well as assess vascular amyloid after di-siRNA treatment?

APP was not affected by di-siRNA treatment. However, lysosomal Aβ aggregation decreased by approximately 50 to 60 percent and extracellular Aβ decreased by approximately 80 to 90 percent. The sustained APP levels may be due to the intracellular burden of Aβ. Also, the Khvorova group did not assess reductions in vascular amyloid compared to the brain. However, when APOE was silenced in the liver, no impact was observed on plaques in the brain.

When should APOE suppressive treatments be administered to maximize benefits and minimize harm?

Early treatment may prevent overexaggerated interferon signaling from activated microglia due to the lower plaque burden compared to individuals in late-stage AD, suggesting that interferon signaling modulation may not be necessary when treating individuals in early stages of disease. However, late treatment may still be beneficial because it can stimulate the clearance of plaques, although interferon response may require modulation. AD may reflect an accelerated model of aging, and many individuals may benefit from early plaque clearance treatment. However, pharmaceutical companies have resisted the idea of silencing APOE, so a white paper from experts in the field may influence pharmaceutical companies to reevaluate the idea.

Combination Therapy in NACC and ADNI Alzheimer’s Participants: Impact of APOE Genotype and Sex

Francesca Vitali, Ph.D., University of Arizona

Multiple AD risk factors, such as diabetes, hypertension, and hyperlipidemia can contribute to and accelerate the onset of AD, and thus addressing these conditions may impact AD progression. Dr. Vitali’s group initially interrogated UK Biobank data to assess the impact of major AD risk factors and their duration stratified by APOE genotype. As expected, APOE4 carriers showed accelerated AD onset compared to non-APOE4 carriers. In addition, some risk factors (e.g., hypertension, obesity) showed greater acceleration of AD onset than others (e.g., depression) independent of APOE genotype.

Next, Dr. Vitali’s group explored whether currently FDA-approved therapeutics for treating AD risk factors can delay or treat AD. The group developed a Targeted-Risk-AD-Prevention approach, in which they identified 629 medications that target one of 364 conditions that are considered AD risk factors from several medical informatics datasets. Based on text mining of literature available on the identified medications and AD, confidence and relevant scores were calculated for each medication’s ability to reduce AD risk factors and identified 46 highly relevant drugs, including hormone therapies, lipid-reducing therapies, anti-inflammatory drugs, and metabolic drugs. The group then performed a pathway analysis of each drug target and found that the 46 medications share many targets, but also have unique targets, suggesting that combination therapies may be more beneficial for treating AD. Furthermore, many medications within the same class also have different targets, suggesting that precision therapeutics may be beneficial for treating AD, such as APOE-targeting medications.

After identifying relevant medications, Dr. Vitali’s group investigated the rate of progression of cognitive decline in real-world clinical data by evaluating cognitive outcomes measures (i.e., the Mini-Mental State Examination [MMSE] and Clinical Dementia Rating Sum of Boxes [CDR-SB]) from 7,653 AD participants from NACC datasets and 411 AD participants from Alzheimer’s Disease Neuroimaging Initiative (ADNI) datasets, all of whom had medication records and at least two clinical visits. The group assessed the impact of diabetes (DBMD), lipid lowering (LIPL), antihypertensive (AHTN), and non-steroidal anti-inflammatory (NSD) medications in AD participants. Findings from these analyses indicated that these drugs, either individually or in combinations, significantly delayed cognitive decline. In particular, NACC participants prescribed the quadruplet of DBMD+LIPL+AHTN+NSD exhibited a 47% delay in MMSE cognitive decline and a 33% delay in CDR-SB cognitive decline at 10 years compared to non-treated participants after adjusting for baseline scores. The ADNI data validated NACC findings, with quadruplets showing 60% delay in cognitive decline for both MMSE and CDR-SB scores at 2 years compared to non-treated participants.

Analysis of APOE genotype differences showed that NACC APOE4 carriers treated with only one medication showed similar cognitive decline to both APOE4 and non-APOE4 carriers that were not treated with medications, as measured by CDR-SB scores. However, cognitive decline was more delayed for APOE4 carriers treated with multiple medications compared to those treated with one medication. The magnitude of cognitive decline delay from both NACC and ADNI quadruplets was slightly greater than the magnitude of delay than participants treated with approved amyloid antibodies. In future efforts, the group plans to investigate the impact of specific therapies within drug classes in AD participants and include analyses of blood-based biomarkers and neuroimaging data.

Discussion

What was the stage of AD among the participants analyzed?

The Vitali group curated ten years of longitudinal data on these participants, in which baseline cognitive scores and recruitment age were curated. The average baseline score of MMSE was 21 and the average baseline score of CDR-SB was 5.

Did analyses adjust for any confounding factors to shift from correlation to future prediction of interventions?

All analyses adjusted for baseline cognitive scores and recruitment age.

Is assessment of the mechanistic components involved in each combination therapy possible? Which agent of each combination is the primary driver of therapeutic benefit?

In the curated data, participants were assigned exclusively to one AD risk factor group, so assessing overlapping AD risk factors was not possible.

Did analyses find any significant differences in medication effects by sex?

No analyses found any significant differences between sexes for any given medication.

Did analyses examine social determinants of health and education levels in treated and non-treated participants?

Analyses did not include education level.

Is it possible that the non-treated participants have untreated co-morbidities that may be influencing your group’s results?

Two controls groups were assessed; one group of non-treated AD participants and one group of non-treated AD participants without any diagnosis of conditions that we selected for analysis.

How can analyses address the possibility that participants treated with multiple medication have greater vascular and less neurodegeneration burden?

Stratifying analyses by vascular dementia and other dementias to assess any differences could help determine vascular and neurodegeneration burden.

Using Biomarkers in Persons with Different APOE Variants to Inform the Study, Treatment and Prevention of AD

Eric Reiman, M.D., Banner Health

Neuropathological and biomarker measurements can inform the impact of APOE4 and APOE2 allelic dose risks of AD. When eliminating clinically characterized cases of misdiagnosis (i.e., cognitively unimpaired individuals), the impact of APOE genotypes on AD development accounts for approximately 25 percent of individuals with mild to moderate AD dementia and 34 percent of APOE4 non-carriers. Furthermore, APOE2 homozygotes showed a 0.004 OR of AD dementia compared to APOE4 homozygotes based on both clinical and neuropathological diagnostic criteria, which suggests the potential for developing APOE-modifying treatments. These OR studies of neuropathological biomarkers also can show the residual effects of APOE genotypes on tau pathology after controlling for amyloid pathology. However, neuropathological studies are often limited in cohort size and the number of underrepresented participants. Although prior studies have analyzed APOE4 differential impact on AD dementia development across ethnic and racial groups, these studies do not assess whether decreased APOE4 impact in some groups existed due to lower risk of APOE4 on AD-specific pathology or lower risk of APOE4 on downstream biomarker effects.

Over the last 30 years, retrospective studies have consistently shown that the risk for developing AD in APOE4 carriers is high. For example, studies of clinically and neuropathology characterized AD cases and controls found that 90 percent APOE4 homozygotes eventually met AD criteria. In addition, 75 percent of APOE4 homozygotes had positive amyloid PET scans and nearly all had low CSF Aβ42 levels by age 65, 88 percent of deceased homozygotes met neuropathological criteria by age 88, and CSF and plasma pTau elevations preceded symptom onset by 10 to 15 years. However, recent prospective cohort studies showed that APOE4 homozygotes showed a lower likelihood of developing cognitive impairment compared to retrospective studies.

From prospective analysis of the Alzheimer's Prevention Initiative Generation Program data, Dr. Reiman’s group found that only 64 percent of cognitively unimpaired 60- to 75-year-old APOE4 homozygotes showed positive amyloid PET scans. From the APOE4 Allelic Dose Study, amyloid PET measurements were lower in APOE4 homozygotes after age 70, suggesting that those who remained unimpaired at older ages have protective factors. Using longitudinal data from cognitively unimpaired 60- to 75-year-old individuals from additional prospective cohort studies, the group found that the risk of MCI or dementia by age 85 was 30 to 55 percent in APOE4 homozygotes, 20 to 25 percent in heterozygotes, and 10 to 15 percent in non-carriers. Within the APOE4 Allelic Dose Study, only one APOE4 homozygote developed dementia by age 70. Together, these findings suggest that larger longitudinal population-based cohort studies stratified by each APOE genotype are needed to clarify the risk of biologically defined AD and ensuing cognitive impairment in unimpaired individuals, including underrepresented groups.

The group recently found that an APOE3 Ch carrier overproduced amyloid for over 30 years compared to other APOE3 carriers in their cohort, but showed protective effects as measured by neurodegenerative biomarkers. These findings emphasize the importance of using blood-based biomarkers in prospective cohort studies. Blood-based biomarkers also present an opportunity to identify effective primary and secondary preventative drug therapies, as well as the role of APOE4 in older adults who are positive for AD biomarkers before the onset of symptoms. Furthermore, researchers can use biomarkers in APOE4 gene-silencing clinical trials to confirm or refute the hypothesis that APOE variants contribute to AD via GOF toxicity.

Discussion

What are the protective factors for APOE4 homozygotes who do not develop cognitive impairment in old age? Given that GWAS studies have identified protective APOE4 variants, how can protective factors be identified?

The Reiman group is currently conducting a study of 300 cognitively unimpaired individuals, of which half are older adults between 70 and 90 years old, across each of the six common APOE genotypes. Deep phenotyping of these individuals is being conducted, but the small sample size presents analysis challenges. Analyses will implement complementary datasets from highly phenotyped cohorts, as well as datasets from poorly phenotyped cohorts that have legacy blood samples available. With these samples, a range of protective factors can be evaluated. In addition, a researcher from Reiman’s group oversees the Alzheimer's Prevention Registry's GeneMatch program, in which approximately 400,000 individuals are registered and receive updates about the latest opportunities to participate in studies. Information about ongoing studies can be sent to individuals characterized by APOE genotype from that registry, but curating a sufficient sample size remains challenging.

Therapeutic Correction of ApoE4-Mediated Endolysosomal Dysfunction in Alzheimer’s Disease

Joachim Herz, M.D., University of Texas Southwestern Medical Center

Along with the common plaque and NFT hallmarks of AD, a recently identified hallmark that manifests in neuronal cells is enlarged endosomes, which occurs in early onset forms of AD mutations such as APOE4. Dr. Herz’s group explored the mechanism by which APOE4 protein enlarges endosomes. The group found that both APOE3 and APOE4 have similar binding affinities for cell-surface receptors, and both APOE-receptor complexes are efficiently endocytosed by endosomes. However, APOE3 protein was released back into cell culture medium, and the receptors recycled back to the plasma membrane, whereas APOE4 protein was retained in early endosomes for an extended period of time. When receptors and co-receptors were trapped in early endosomes due to APOE4 aggregation, a fraction was degraded in the endolysosomal compartments rather than trafficked back to the plasma membrane. When analyzing the isoelectric point (IEP) of the APOE protein, the group found that APOE2 has the most acidic IEP at a pH of ~5.9, and APOE4 has the most basic IEP at a pH of ~6.4. Moreover, the IEP of APOE4 approximately matched the pH of early endosomes.

Prior literature describing the purification of insulin in the 1920s showed that proteins tended to self-aggregate and lose solubility in environments matching their own IEP. Thus, Dr. Herz’s group hypothesized that this APOE self-aggregation mechanism was extending the lifetime of early endosomes and preventing trafficking maturation of this cellular compartment. Assuming this hypothesis holds true, the group further hypothesized that lowering the pH of early endosomes can resolve these dysfunctions, which can be achieved by blocking proton leak in endosomes with a sodium-hydrogen exchanger (NHE) 6 inhibitor or enhancing proton pump activity. For example, a prior study showed that administration of the miRNA miR-1 in aged Caenorhabditis elegans (C. elegans) models increased endosomal proton pump activity and subsequently increased proteotoxic clearance. In addition, the group proposes that the rate of endosomal proton gradient deterioration may determine the age-of-onset of AD symptoms. This is evident by studies showing that in early onset disease animal models, endosomal pH increases. This suggests that conversely in centenarians loss of endosomal pH gradient may be delayed. Thus, balancing blocking sodium-hydrogen exchanger activity and aiding proton pump activity may be a viable therapeutic strategy for treating AD, not only in APOE4 carriers but in general.

Discussion

In this study, does miR-1 activate a transcription factor to enhance proton pump activity?

That study identified a miRNA that regulates the expression of C. elegans vacuolar ATPase proton pump subunits. During C. elegans aging, a reduction of endolysosomal esterification occurs, which allowed the investigators to test an antagonist against the miRNA. The antagonist successfully suppressed the miRNA’s activity, so the miRNA failed to suppress the transcription factor that regulates the expression of vacuolar ATPase.

Do GWAS studies identify any genes associated with changes in endosomal proton gradient degradation?

The NHE6 proton pump regulates the pH levels of early endosomes and is encoded by an X-chromosomal gene. Because only one functional copy per cell exists, exploring whether heterozygotes carrying an NHE6 variant are protected against AD is not possible. But based on conversations with Dr. Greicius, there is no statistical evidence of NHE6 variants that protect against AD.

Could acidifying endosomes potentially contribute to the clearance of amyloid as well?

The hypothesis was based on the self-aggregation of APOE protein, so we believed that acidifying endosomes would primarily aid in the restoration of endosomal recycling and intracellular trafficking. The Herz group conducted an experiment in which non-variant APOE mouse models were used as controls, and the effect on amyloid clearance was similar to the hypothesis about APOE4 endosomal clearance. Furthermore, a recent study showed that decreased endosomal acidification occurs in all AD mouse models, including the mouse model our group used. Those findings provide evidence that increasing endosomal acidification can also help amyloid clearance. Impaired acidification may also be a primary underlying mechanism of AD; underlying endosomal transport and sorting mechanisms may be impaired, leading to amyloid accumulation.

Does modifying endosomal acidification with reagents affect intracellular neuronal environments, extracellular environments, or both? If both are impacted, could this acidification impact normal neuronal activity?

NHE6 dysfunction can cause Christianson syndrome, which leads to hyperacidification of early lysosomal compartments and has severe developmental consequence due to premature degradation of neurotrophin receptors. However, the mechanisms underlying Christianson syndrome should not deter the development of NHE6 inhibitors because during aging, individuals do not exhibit hyperacidification of lysosomal compartments, but rather hypoacidification. Thus, partial inhibition of primary proton leak channels may be beneficial.

In which cell types could APOE4 aggregation be more effectively eliminated by the acidification of endosomes?

Endosomal acidification occurs in all cells, including hepatocytes and neuronal cells, which may explain why lower plasma APOE levels are correlated with increased AD risk. APOE4 protein remains in lysosomal compartments for an extended period of time, and a subsequent fraction of APOE4 protein is diverted to lysosomal degradation, both of which may result in lower plasma APOE4 levels.