The Longevity-Linked APOE2 Gene Supports DNA Signals That Help Human Neurons Resist Aging

To determine how APOE2 contributes to neuronal resilience in inhibitory neurons, we differentiated human female isogenic iPSCs carrying homozygous APOE2 or APOE4 alleles into GABAergic neurons. Because interneuron dysfunction is implicated early in AD and APOE4 expression in interneurons suggests vulnerability to synaptic deficits and connectivity, we focused on this neuronal subtype vulnerability in neuronal connectivity (Knoferle et al. 2014; Najm et al. 2019). Immunocytochemistry confirmed robust expression of VGAT (vesicular inhibitory amino acid transporter) and GABA (γ‐aminobutyric acid)—markers of neurotransmitter synthesis and vesicular transport (Chaudhry et al. 1998)—as well as the neuronal markers CALBINDIN and NESTIN (Kojetin et al. 2006; Steinert et al. 1999) (Figure 1a). In agreement with previous reports (Brecht et al. 2004; Knoferle et al. 2014; Najm et al. 2019), APOE4 neurons exhibited a pronounced loss of GABAergic neurite network density, relative to APOE2 neurons, over a 21‐day culture period (Figure 1b). Quantitative live‐cell imaging revealed a significantly decreased neurite complexity in APOE4 neurons at 300, 450, and 600 h of differentiation. Beyond neurite architecture, genotype‐specific alterations in neuronal motility were observed. Across 21 days of live imaging, APOE4 neurons displayed increased mean curvilinear velocity but reduced path velocity, along with greater wobble amplitude compared to APOE2 neurons (Figure 1c, Figure S1a). These findings reveal that APOE2 and APOE4 GABAergic neurons exhibit distinct cellular phenotypes, highlighting divergent effects of APOE genotype on inhibitory neuron structure and dynamics.

To understand the distinct cellular phenotypes and underlying molecular mechanisms of the APOE isoforms, we performed an unbiased transcriptomic analysis of APOE2 and APOE4 GABAergic neurons. Principal component analysis (PCA) showed that the APOE2 and APOE4 GABAergic neurons segregated into distinct clusters (Figure 2a). We identified 1403 differentially expressed genes (DEGs) between APOE2 and APOE4 neurons (Figure 2b,c, Table S1). APOE mRNA was expressed in both genotypes (Figure S1b).

Among the most upregulated genes (APOE2 vs. APOE4) were DSCAM (Down syndrome cell adhesion molecule), which regulates neuronal delamination (Arimura et al. 2020) and locomotion (Lemieux et al. 2021), and CSMD2 (CUB and Sushi multidomains 2), which is involved in the development and maintenance of dendrites, synapses, and brain structure (Gutierrez et al. 2019). Neuronatin (NNAT), a gene upregulated in AD and associated with spine loss and endoplasmic reticulum calcium overload (Zou et al. 2022), was among the most downregulated genes in APOE2 neurons. Contactin‐4 (CNTN4), involved in cell adhesion, APP function, and AD (Bamford et al. 2024), was also reduced in APOE2 neurons (Figure 2b,c).

Gene Ontology (GO) analysis revealed that genes upregulated in APOE2 neurons were significantly enriched for pathways related to DNA damage response and repair, whereas genes upregulated in the APOE4 genotype were enriched for pathways involving synaptic activity, axonogenesis, and ion channel function (Figure 2d; Table S1). Network analysis of the DNA repair signaling pathway identified BRCA1 (breast cancer 1), CDK1 (cyclin‐dependent kinase 1), PLK1 (polo‐like kinase 1), and TOP2A (DNA topoisomerase II alpha) as central hub genes in the APOE2 GABAergic neurons (Figure 2e). A heatmap of the DNA damage checkpoint genes further confirmed upregulation of factors involved in DNA repair, damage resistance, and genomic stability, including BLM (Bloom syndrome recQ‐like helicase), RAD9B (checkpoint clamp component B), BRCA1, and PLK1 (Figure 2f) in APOE2 GABAergic neurons. Together, these results indicate allele‐specific transcriptional programs, with APOE2 associated with enhanced DNA damage response mechanisms, and APOE4 neurons biased toward synaptic signaling programs.

To assess genotype‐specific transcriptional differences at single‐cell resolution, we performed 10x Genomics single‐cell RNA sequencing (scRNA‐seq) on GABAergic neurons derived from APOE2 and APOE4 backgrounds. After quality filtering (e.g., retaining cells with ≥ 300 features, > 1000 unique molecular identifiers (UMIs), and < 25% mitochondrial reads), transcriptomes from 7328 cells were retained for analysis. Using the Seurat analysis pipeline, we identified seven transcriptionally distinct clusters, with cells from both genotypes across all clusters (Figure 3a) (Satija et al. 2015).

Although both genotypes contributed to each cluster, their relative distributions differed. Clusters 1, 3, and 4 were enriched for APOE4 neurons, whereas clusters 0, 2, and 5 were more enriched in the APOE2 neurons (Figure 3b,c). Cluster‐specific marker genes were identified by differential expression analysis (Figure 3d) and visualized using violin plots (Figure S2a).

APOE2 neurons have higher expression of VGF (nerve growth factor inducible) in Cluster 0 than APOE4 neurons (Figure 3d). VGF is a neuropeptide precursor downregulated in AD (El Gaamouch et al. 2020), a top causal master regulator of AD networks (Beckmann et al. 2020), and overexpression of VGF rescues phenotypes in AD mice (El Gaamouch et al. 2020). Cluster1 (Figure 3e, Table S2, volcano plot E2/E4 comparison), enriched in APOE4 neurons, has higher expression of VCAN (Versican), TXNIP (thioredoxin‐interacting protein), and VSNL1 (Visinin‐like 1, aka VILIP‐1, HLP3). VSNL1 is a neuronal calcium sensor protein and is increased in the hippocampus and entorhinal cortex, which are affected early in AD (Groblewska et al. 2015; Halbgebauer et al. 2022; Tarawneh et al. 2015). Cluster 2, enriched for APOE2 neurons, has high expression of neuronal genes KCNMB2 (potassium calcium‐activated channel subfamily M regulatory beta) (Bentrop 2001) and IGSF21 (immunoglobulin superfamily member 21). IGSF21 is expressed on the postsynaptic membrane and stabilizes inhibitory synapses (Tanabe et al. 2017).

To understand functional divergence among clusters, we performed gene set enrichment analysis (Figure S2b). Core “hallmarks” shared across clusters included TNF‐alpha signaling, hypoxia, and inflammatory pathways. Interestingly, cluster 0, enriched in APOE2 neurons, showed specific enrichment for the mitotic spindle and in DNA damage response to UV light. Overall, pathway enrichment revealed a distribution of 24% inflammatory signaling and 19% DNA repair pathways (Figure S2c).

Given the strong transcriptomic signature of APOE2 GABAergic neurons enriched in DNA signaling, we assessed relevant phenotypes associated with DNA damage. Immunocytochemistry was performed on the isogenic APOE2 and APOE4 GABAergic neurons using the DNA‐damage repair‐associated protein p‐γH2AX (Figure 4a), which correlates with double‐stranded breaks and residual DNA damage (Ivashkevich et al. 2012; Paull et al. 2000; Sharma et al. 2012). p‐γH2AX staining was greater in APOE4 than in APOE2 GABAergic neurons (Figure 4b). To determine if the APOE4 neurons undergo increased DNA damage rather than just altered repair pathways, we performed the comet assay to analyze DNA breaks (Larson 2016; Olive and Banath 2006) (Figure 4c). We examined the tail DNA percent (percentage of tail DNA over DNA of the entire cell) and the olive tail moment of the comets. The APOE4 neurons had a greater percentage of DNA in their tails than APOE2 neurons. The olive tail moment value parameter, a composite parameter reflecting DNA migration distance and distribution, was also greater for APOE4 than for APOE2 (Figure 4d). Thus, more DNA damage was found in APOE4 than in APOE2 GABAergic neurons.

Aberrant expression of repetitive elements (e.g., retrotransposons) is a hallmark of aging and neurodegenerative diseases (De Cecco et al. 2019; Saleh et al. 2019). These repetitive elements are often associated with hot spots of DNA damage (Argueso et al. 2008). To determine if the APOE genotypes of GABAergic neurons changed the expression of repetitive elements, we used RepEnrich2 to quantify the repetitive elements in our RNA‐seq data. We also used DESeq2 to identify differentially expressed repetitive elements (Table S3) (Love et al. 2014). PCA of the repetitive elements showed that APOE2 and APOE4 GABAergic neurons were clustered separately. We analyzed the composition of the repetitive element landscape, based on class membership, which includes repetitive rRNA (ribosomal RNA), LINE, SINE, snRNA, and LTR elements (Figure 4e). APOE4 GABAergic neurons had greater rRNA expression and lower expression of SINE, LINE, and LTR elements (Figure 4e). Overall, our results suggest that APOE2 neurons are more resistant to DNA damage and do not exhibit the increased rRNA expression observed in APOE4 neurons. APOE4 neurons have greater rRNA expression, which has been associated with cellular senescence and nucleolar stress (Morlot et al. 2019).

APOE2 GABAergic neurons show upregulation of categories related to DNA repair and stability, suggesting enhanced genomic maintenance capacity and relative resistance to DNA damage. Moreover, these neurons resist an increase in repetitive elements, another hallmark of aging associated with cellular senescence. Therefore, we evaluated whether APOE2 might confer protection against cellular senescence and DNA damage induced by genotoxic stress using a second neuronal model—excitatory neurons—with all three human APOE genotypes. Using Cas9/CRISPR‐engineered isogenic male iPSCs with APOE alleles (APOE ε2/ε2, ε3/ε3, and ε4/ε4 genotypes), the iPSCs were differentiated into excitatory glutamatergic neurons using inducible expression of Neurogenin‐2 (Ngn‐2). After 28 days of differentiation, the cells expressed PSD95 (postsynaptic density protein‐95) and VGLUT1 (vesicular glutamate transporter 1), markers of mature glutamatergic neurons, as well as MAP2 (Microtubule‐associated protein 2), a pan neuronal marker expressed across all genotypes (Figure S3a). Neurons of all APOE genotypes expressed APOE as confirmed by immunocytochemistry and western blot analysis (Figure S3b,c). Subsequently, glutamatergic neurons were exposed to irradiation (10 Gy) or doxorubicin (200 nM) to evaluate key markers associated with DNA damage and cellular senescence (Figure 5a).

Long‐lived animals have decreased expression of rRNA and smaller nucleoli (Tiku et al. 2017), consistent with our previous observation of reduced rRNA levels in the APOE2 GABAergic neurons. The nucleolus is a nuclear subcompartment for rRNA synthesis and ribosomal subunit assembly and plays a crucial role in genome integrity, nuclear architecture, stress signaling, and cell‐cycle regulation. Reduced nucleolar size has been associated with longevity and metabolic health (Tiku et al. 2017). Therefore, we evaluated the effect of APOE isoforms on the nucleolar size by immunostaining against Nucleolin (NCL), a protein involved in the synthesis of rRNA. Strikingly, APOE2 and APOE3 glutamatergic neurons displayed significantly smaller nucleoli than APOE4 neurons under basal conditions and irradiation‐induced senescence (Figure 5b). Thus, the nucleolar size of APOE2 glutamatergic neurons aligns with a cellular feature associated with longevity and metabolic health.

Cellular senescence, an important process in aging, is characterized by the upregulation of p16/CDKN2A and p21/CDKN1A, DNA damage response, nuclear enlargement, chromatin reorganization, and the expression of pro‐inflammatory factors. The morphological alterations of the nucleus can serve as a predictor of senescence (Heckenbach et al. 2022). Therefore, we evaluated the nuclear size of the APOE glutamatergic neurons. Following irradiation‐induced senescence, the nuclear size of glutamatergic neurons increased in all genotypes compared to control conditions. The nuclear size of APOE2 neurons was smaller than that of APOE4 (Figure S4a). To determine whether APOE isoforms are differentially susceptible to genotoxicity‐induced cellular senescence, we measured p16 expression. Irradiation significantly induced p16 in all genotypes compared to the control conditions. Notably, APOE4 neurons showed higher levels than APOE2 and APOE3 (Figure 5c).

We further assessed CRYAB (αB‐crystallin), a small heat shock protein, which is upregulated in senescent cells and can be used as a senolytic target (Limbad et al. 2022). Notably, APOE2 neurons displayed a lower percentage of CRYAB‐positive cells than APOE4 under basal conditions. Likewise, CRYAB‐positive cells increased after irradiation‐induced senescence in all genotypes, compared to controls, but APOE4 neurons contained a higher proportion of CRYAB‐positive cells than APOE2 and APOE3. Interestingly, more CRYAB‐positive cells were in APOE4 neurons under basal conditions than in APOE2 neurons (Figure 5d). Altogether, these results suggest that APOE2 glutamatergic neurons are resilient to genotoxicity‐induced senescence and that APOE4 neurons are more prone to acquire a senescent‐associated phenotype.

As described above, APOE2 GABAergic neurons exhibited upregulation of pathways related to DNA damage response and repair. We therefore evaluated whether APOE2 confers protection against DNA damage and resistance to genotoxicity‐induced senescence following irradiation or doxorubicin treatment. DNA damage accumulates with age due to increased reactive oxygen species production and declining DNA repair, leading to DNA damage response and cellular senescence. Irradiation significantly increased p‐γH2AX, 53BP1, and p‐ATM across all genotypes, compared to control conditions, consistent with activation of DNA damage response, a key driver of senescence. Notably, APOE4 glutamatergic neurons exhibited a higher number of p‐γH2AX foci per nucleus, as well as larger foci, compared to APOE2 under irradiation‐induced senescence (Figure 5e). Similarly, 53BP1 and p‐ATM analysis revealed that following irradiation, APOE4 neurons had increased foci size compared to APOE2 and APOE3 neurons (Figure 5f,g). The statistically significant change in p‐ATM foci size during irradiation was more pronounced in APOE3 and APOE4 neurons relative to APOE2 neurons. Interestingly, the basal levels of 53BP1 mean intensity were APOE allele‐dependent (Figure S4b).

Correspondingly, in doxorubicin‐induced cellular senescence, APOE2 neurons had lower levels of p16 than APOE3 and APOE4 neurons (Figure S5a). Doxorubicin significantly increased p‐γH2AX, 53BP1, and p‐ATM across all genotypes compared to control conditions. Moreover, APOE4 neurons displayed greater p‐γH2AX numbers of foci per nucleus, larger 53BP1, and p‐ATM than APOE3 and APOE2 (Figures S5b and S6a,b). Collectively, these findings indicate APOE2 glutamatergic neurons are resilient to DNA damage after genotoxic stress.

To evaluate early DNA damage response dynamics, we performed a time course analysis in APOE neural stem cells (NSCs). Consistent with the genotype‐specific response to DNA damage observed in differentiated neurons, p‐γH2AX, p‐ATM, and 53BP1 resolved more rapidly in APOE2 NSCs compared to APOE4 NSCs (Figure S7). These results suggest that APOE2 promotes a more efficient DNA damage response.

Given that DNA damage can lead to epigenomic alterations and nuclear instability, we examined the nuclear lamina integrity and heterochromatin status. Loss of nuclear lamina components and heterochromatin is associated with genomic instability and aberrant transcription during cellular senescence. Therefore, we evaluated LAMIN A/C, a major component of the nuclear lamina, as well as H3K9me3, a heterochromatin‐associated histone modification linked to chromatin organization and cellular senescence. Under basal conditions, APOE3 and APOE4 neurons displayed lower LAMIN A/C levels compared to APOE2 neurons (Figure 5h). Notably, APOE2 neurons were resistant to irradiation‐induced downregulation of LAMIN A/C observed in APOE3 and APOE4 neurons (Figure 5h). In contrast, APOE4 neurons exhibited higher basal H3K9me3 levels compared to APOE2 and APOE3 neurons. Notably, H3K9me3 levels decreased in APOE3 and APOE4 neurons relative to control conditions but not in APOE2 neurons (Figure S4c). These findings may suggest that APOE2 glutamatergic neurons maintain greater nuclear and chromatin stability following genotoxic stress.

As described above, APOE2 GABAergic neurons exhibited reduced DNA damage, and APOE2 glutamatergic neurons were more resilient to genotoxic‐induced DNA damage. We therefore tested whether recombinant APOE2 treatment could protect APOE4 glutamatergic neurons from irradiation‐induced DNA damage. Recombinant APOE2 treatment reduced p‐γH2AX and 53BP1 number of foci per nucleus and foci size following irradiation (Figure 6a). These results indicated that APOE2 attenuates irradiation‐induced DNA damage signaling in APOE4 neurons.

Next, we evaluated age‐related markers in brain tissue from 16‐month‐old APOE knock‐in mice. We focused on the dentate gyrus, a region critically involved in learning and memory and vulnerable in AD. APOE2 and APOE3 knock‐in mice displayed smaller nucleoli compared to APOE4 mice (Figure 6b), consistent with the association between nucleolar enlargement and aging. We next assessed the nuclear lamina integrity by measuring Lamin A/C levels. APOE2 knock‐in mice had higher levels of Lamin A/C than APOE3 and APOE4 mice (Figure 6b), consistent with preserved nuclear structure. We also evaluated H3K9me3 levels, a heterochromatin‐associated histone modification linked to chromatin organization and aging. Aged APOE4 and APOE3 mice exhibited reduced levels of H3K9me3 in the dentate gyrus compared to APOE2 mice (Figure 6b). In addition, we quantified nuclear Hmgb1 levels, as Hmgb1 translocation to the cytosol is associated with senescence. Aged APOE2 mice retained higher nuclear Hmgb1 levels compared to APOE3 and APOE4 mice (Figure S8a). Finally, using publicly available RNA‐seq data from 18‐month‐old APOE mice, we performed enrichment analysis for the Hallmarks of Aging gene sets (Labuza et al. 2025). Gene sets related to cellular senescence were significantly enriched in APOE4 compared to APOE2 mice (Figure S8b). Collectively, these findings support the conclusion that APOE2 is associated with molecular features indicative of healthier aging in vivo.

Human APOE4 GABAergic neurons (Fujifilm, R1168, 4 million cells per vial) and APOE2 GABAergic neurons (Fujifilm, R1169, 4 million cells per vial) were obtained from Fujifilm (formerly Cellular Dynamics International). Cells were thawed and maintained according to the manufacturer's instructions using iCell Neural Base Media 1 and its associated Supplement A (R1150), which together constitute the Complete Maintenance Medium. For culture, 8‐well chamber slides, 12‐well plates, 6‐well plates, and/or 96‐well plates were coated with one of two matrix conditions. The first condition, plates were coated with 0.01% poly‐L‐ornithine (Sigma‐Aldrich, P4957) and left overnight at 37°C, rinsed 3× each with Milli‐Q water (Elga Superflex Model PF3), and then coated with 3.3 μg/mL laminin (Sigma‐Aldrich, L2020) in Milli‐Q water for 1 h at 37°C. Wells were subsequently filled with Complete Maintenance Media. In the second condition, plates were coated with 0.1 mg/mL poly‐D‐lysine (Sigma‐Aldrich, P6407) overnight at 37°C, rinsed three times with Milli‐Q water, and then coated with Matrigel (130–160 μg/mL, Corning, CB40234) overnight at 37°C. Wells were aspirated and replenished with SynaptoJuice B as described in (Kemp et al. 2016). For most experiments, SynaptoJuice B was used, as the GABAergic neurons exhibited improved viability and morphology in this medium. Media were changed every 3–5 days. The cells were allowed to mature for at least 14 days in culture, and 2–5 batches were used depending upon the experiment.

APOE4 and APOE2 GABAergic inhibitory neurons were cultured as described above and imaged in brightfield every 3 h using a BioTek Cytation 5 instrument at 4× or 10× magnification. Image acquisition and analysis were performed using Gen5 software (BioTek, Version 3.08). Images were preprocessed using background correction and deconvolution (3× iterative cleaning step) to reduce optical blur and enhance neurite visibility. Processed images were used for neurite outgrowth quantification and for the generation of time‐lapse videos for cell motility analysis. Neurite outgrowth was quantified by applying a primary mask to processed images, with a mask threshold of 750 units. The “Sum Perimeter” (SP) parameter, defined as the total perimeter of all detected objects within the field of view, was extracted for each image. Objects within an acceptable size range of 5–500 μm were included in the analysis. Once optimized, mask parameters were applied uniformly across all wells and time points. To normalize for baseline variability, SP values at each time point were divided by the SP value of the first time point (read 1) for the corresponding well, generating a normalized Sum Perimeter (SP_norm). Temporal changes in SP_norm were plotted using GraphPad Prism 8.

GABAergic neurons were cultured and imaged at 4× magnification as described above. For each well, 2 × 2 tiled images centered within the well were acquired and stitched using BioTek software. Time‐lapse image sequences were registered and exported as H.264‐compressed (MP4) videos. Cell body motility was analyzed using Image Analyst MKII (Image Analyst Software, Novato, CA). Cell bodies were identified in brightfield images using band‐pass spatial filtering with absolute value transformation, followed by image segmentation based on a modified version of the “Measure tracking parameters of cells in brightfield time lapses” pipeline. Individual cells were tracked across frames using a simulated annealing optimization algorithm implemented in the “Track Objects” function, which associates objects between consecutive frames based on proximity, size, and shape features without requiring direct spatial overlap. Tracks shorter than 10 frames were excluded from analysis. Because cell movement in culture is largely random, the vector sum of all displacement vectors within a field of view was assumed to be zero; any residual displacement was attributed to imperfect frame registration and was subtracted from each frame‐to‐frame displacement vector. Erroneous tracking events were excluded using a velocity threshold cutoff. The remaining tracks were quantified using the “Plot Tracking Parameters” function. Mean curvilinear velocity (VCL) was calculated as the average absolute velocity along the entire track. Average path velocity (VAP) was determined by smoothing velocity vectors prior to calculating absolute velocity, thereby reducing values for trajectories with frequent directional changes while preserving velocities of straight‐moving cells. Wobble was calculated as:. A value of 0 indicates smooth, linear motion, whereas higher values reflect increased directional variability. Tracking parameters were calculated for each cell and subsequently averaged across fields and wells. Wobble = − 1 VAP VCL

GABAergic inhibitory neurons were cultured for 6–10 days on 8‐well chamber‐slides (BD Falcon, 354108), fixed with 4% paraformaldehyde (Sigma, 158127) for 15 min at RT and washed 3× with PBS. Neurons were permeabilized with 0.1% Triton X‐100 (Fisher Scientific, BP151‐100) in PBS solution for 15 min, washed with PBS, and then blocked 30–60 min with 1% BSA (Sigma Aldrich, 03117332001) and 5% donkey serum (Sigma Aldrich, D9663) in PBS. An additional PBS wash was performed, and the cells were probed with primary antibodies overnight at 4°C. Neurons were washed 3× with PBS, and then probed with antigen‐matched, fluorescent‐labeled secondary antibodies for at least 1.5 h and washed 3× times with PBS in the dark. The cells were mounted with 12‐mm glass coverslips (48393‐251) using Prolong Gold with DAPI (ThermoFisher, P36931↗). Imaging was done using a Zeiss LSM 780 confocal on an Axio Observer Z1 inverted microscope with Plan‐Apochromat 63×/1.40 NA Oil DIC objective. Primary antibodies (1:100 dilution) include: VGAT (Synaptic Systems, 131011), GluR1 (Millipore, ABN241), PSD95 (Sigma, SAB5600103), GAD (Millipore, AB1511), phospho‐γH2AX (Millipore, 05‐636), Nestin (Abcam, ab92391), GABA (Synaptic Systems, A2052), MAP2 (Abcam, AB5622), P16 (Abcam, AB108349↗), Calbindin (Abcam, AB108404↗), Nucleolin (Abcam, AB22758), phospho‐ATM (ThermoFisher Scientific, MA1‐2020), 53BP1 (Cell Signaling, 4937), and H3K9me3 (Abcam, AB8898). Secondary antibodies (1:200–1:350 dilution) include: donkey‐anti mouse Alexa Fluor 488 (ThermoFisher, A21202) and donkey‐anti rabbit Alexa Fluor 555 (ThermoFisher, A32794). The phospho‐γH2AX images were analyzed using the Cytation 5 Biotek system quantifying the puncta in APOE4 and APOE2 GABAergic neurons.

Bulk RNA was extracted from 12‐well plates using the Bioline Isolate II RNA mini kit (Bioline, BIO‐52073) and the manufacturer's protocol. Each well contained approximately 400,000 GABAergic inhibitory neurons. The resulting elute was quantified on the NANODROP 2000 (D632) and frozen at −20°C. Three samples of each genotype for six total samples were sequenced by the UC‐Davis core. FASTQ files were assessed for sequencing quality using per‐base quality and adapter content. Reads were trimmed with Trimmomatic to remove adapter sequences and low‐quality bases. Trimmed reads were aligned to the human reference genome, and gene counts were generated by assigning aligned reads to annotated genes. Differential gene expression was performed using DESeq2, and low‐expressed genes were filtered out before testing. p values were corrected for multiple testing using the Benjamini‐Hochberg false discovery rate (FDR).

GO was performed using a ranked list of differential gene expression with parameters set to 2000 gene‐set permutations and gene‐set size between 15 and 200. The gene‐sets included for the Gene Set Enrichment analyses were obtained from Gene Ontology (GO) database (GOBP_AllPathways), updated September 01, 2019 (http://download.baderlab.org/EM_Genesets/↗). An enrichment map (version 3.2.1 of Enrichment Map software (Merico et al. 2010)) was generated using Cytoscape 3.7.2 using significantly enriched gene‐sets with an FDR < 0.05. Similarity between gene‐sets was filtered by Jaccard plus overlap combined coefficient (0.375). The resulting enrichment map was further annotated using the AutoAnnotate Cytoscape App.

Wells from 12‐well plates were harvested for use in single‐cell analysis. Three wells per genotype were harvested, for a total of six samples. We utilized the V3 Chromium Single Cell 3′ Reagent Kits User Guide and its associated Chromium i7 Multiplex Kit (10X, PN‐120262). Data have been deposited in NCBI's Gene Expression Omnibus GEO Series accession number GSE143276↗.

FASTQ files from 10× single‐cell RNA sequencing were processed using Cell Ranger (GRCh38, 2020). Downstream bioinformatics analysis was performed in R using Seurat v4.3.0. Genes detected in fewer than three cells were removed. To filter out cells with low‐complexity libraries, the following thresholds were applied: number of features ≥ 300, log10(GenesPerUMI) > 0.85, mitochondrial RNA < 25%, and ribosomal RNA > 5%. Doublets were identified using DoubletFinder v2.0.6, accounting for 4% of the dataset, and subsequently removed. The novelty score, defined as the ratio of the number of genes to UMIs, was also calculated to ensure that all retained cells exhibited high transcriptomic diversity (novelty score > 0.8). Counts were normalized and scaled with SCTransform. Samples were batch‐corrected using the Harmony method. Unsupervised clustering based on transcriptional data was performed using Seurat's FindNeighbors and FindClusters functions. Clusters were visualized with RunUMAP and DimPlot using default settings. To characterize clusters, we used FindConservedMarkers and FindAllMarkers. Differential expression analysis was performed using a pseudo‐bulk approach and DESeq2 v1.42.1. Gene set and pathway enrichment analyses were conducted using clusterProfiler v4.10.1 and fgsea v1.28.0.

Female homozygous APOE2 (Fujifilm, R1169 donor ID 01434), and APOE4 (Fujifilm, R1168, donor ID 01434) iCell GABAergic neurons were cultured in Corning 6‐well plates (Thermo Fisher Scientific, 140675) coated with 0.1 mg/mL poly‐D‐lysine (Sigma‐Aldrich, P6407) overnight at 37°C. Wells were rinsed 3× with Milli‐Q water and subsequently coated with 130–160 μg/mL Matrigel (Corning, CB40234) overnight at 37°C. SynaptoJuice B medium was added as previously described (Kemp et al. 2016). DNA damage was assessed using the alkaline comet assay (Trevigen/R&D Systems, 4250‐050‐03) according to the manufacturer's protocol. Approximately 3000 cells per condition were collected into 1.5‐mL microcentrifuge tubes (Sorenson, 11590) and mixed with 75 μL of pre‐warmed 1% low‐melting‐point agarose (Bio‐Rad, 1613102). The agarose–cell suspension was evenly spread onto Trevigen 2‐well comet slides and allowed to gel at 4°C for up to 30 min in the dark. Slides were immersed in pre‐chilled lysis solution (Trevigen, 4250‐050‐01) and incubated overnight at 4°C in the dark. For DNA unwinding, slides were transferred to freshly prepared alkaline unwinding solution (8 g NaOH and 2 mL 500 mM in 1 L distilled water; pH ≥ 13) and incubated for 1 h at 4°C in the dark. Slides were then placed in a Trevigen electrophoresis chamber (4250‐050‐ES) containing unwinding solution and subjected to electrophoresis at 21 V for 10 min at 4°C. Following electrophoresis, slides were washed twice with Milli‐Q water for 5 min each, rinsed once with 70% ethanol for 5 min, and air‐dried at 37°C in a carbon dioxide–free incubator for 10 min. DNA was stained with 1× SYBR Gold (Thermo Fisher Scientific, S11494) in Milli‐Q water for 30 min at room temperature in the dark. Slides were rinsed with Milli‐Q water and allowed to dry overnight at room temperature. Comets were imaged using a BioTek Cytation 5 imaging system (Gen5 software, version 3.08) and analyzed as described above.

We used RepEnrich2 to estimate the number of repetitive elements observed in our RNAseq data (Criscione et al. 2014). We used the repetitive element annotation for Homo sapiens from Repeatmasker.org↗. Once repetitive elements were quantified, DESeq2 was used to perform differential enrichment analysis (Love et al. 2014).

The isogenic homozygote APOE2, APOE3, and APOE4 iPSC lines were CRISPR edited using NCRM1 parental clones with Cas9 and gRNA (Synthego, Redwood City, CA). The three lines were fully sequenced. The homozygote lines had the expected sequences and no off‐target edits, and were karyotyped as normal. Human iPSC‐derived neurons were engineered to have a safe harbor locus with integration of a doxycycline‐inducible Ngn2 transgene into APOE2, APOE3, and APOE4 carrying human iPSCs (Wang et al. 2017). The TALENS and donor construct as 1:2 ratio. TALEN AAVS1A (3 μg), AAVS1B (3 μg), and pUC‐hNgn2/oligo (6 μg) with puromycin selection were as described (Wang et al. 2017). All NCRM1 APOE2, APOE3, and APOE4 PCR‐identified clones were karyotyped (Cell Line Genetics) and had normal karyotypes.

We differentiated glutamatergic neurons from human isogenic iPSCs (APOE ε2/ε2, ε3/ε3, and ε4/ε4 genotypes) by inducible Ngn2 expression, using a transgenic cassette that was integrated into the AAVS1 safe harbor locus of each APOE iPSC clone as described (Wang et al. 2017). iPSCs were maintained in colonies in 6‐cm plates pre‐coated with Matrigel (Corning, 354234) using mTESR Plus basal medium (Stemcell, 100‐0274). For differentiation, iPSCs were detached into single cells with StemPro Accutase (Gibco, A1110501) and plated on a Matrigel‐coated plate at a density of 2.5 × 10⁵ cells/cm². iPSCs were maintained using N2+ medium containing KO DMEM/F12 (ThermoFisher, 12660012), 1% N2 supplement (ThermoFisher, 17502001), 1× NEAA (ThermoFisher, 11140050), 10 ng/mL BDNF (Peprotech, 450‐02), 10 ng/mL NT3 (Peprotech, 450‐03), and 2 μg/mL doxycycline. The medium was changed every day. After 3 days of pre‐differentiation, cells were detached using StemPro Accutase and plated in poly‐D‐lysine (20 μg/mL) coated coverslips or plates at a density of 7.5 × 10⁴ cells/cm² and maintained using NBA+ medium containing Neurobasal A (ThermoFisher, 108888022), 2% B27 (ThermoFisher, 17504044), 1x GlutaMAX (ThermoFisher, 35050061), 10 ng/mL BDNF (Peprotech, 450‐02), 10 ng/mL NT3 (Peprotech, 450‐03). Neurons were differentiated for 28 days, and half of the medium was changed every week. A subset of plates were exposed to irradiation (10 Gy) or doxorubicin treatment (200 nM) at day 18 during the differentiation and analyzed 10 days after treatment. For recombinant APOE2 experiments, a subset of plates was treated with recombinant APOE2 (50 ng/mL) starting at day 11 twice a week until analysis.

Differentiated neurons were cultured on coverslips, fixed with paraformaldehyde 4% for 15 min at RT, and stored in PBS 1× at 4°C. Neurons were blocked and permeabilized using 1X PBS, 5% fish gelatin (Sigma, G7041), and 0.5% Triton‐X‐100 (Sigma, X100) for 2 h at RT. Primary antibodies were incubated overnight at 4°C in blocking solution at a concentration of 1:250 for p‐γH2AX (Millipore, 07‐627), p‐ATM (ThermoFisher, MA1‐2020), 53BP1 (CellSignaling, 4937), CRYAB (Abcam, ab13496), H3K9me3 (Abcam, ab8898), 1:750 for p16/cdkn2a (Abcam, ab108349), and MAP2 (Novus Biologicals, NB300‐213). Secondary antibodies were incubated for 2 h at RT in blocking solution at a concentration of 1:500 for anti‐mouse Alexa 488 (ThermoFisher, A32766), anti‐rabbit Alexa 555 (ThermoFisher, A32794), and anti‐chicken Alexa 647 (ThermoFisher, A21443), followed by incubation with DAPI (1 μg/mL) for 10 min at RT. Coverslips were mounted on a slide using ProLong Gold antifade mountant and stored at 4°C in the dark. Images were taken using a confocal microscope ZEISS LSM 980. Images were analyzed using Fiji Image J.

Neural stem cells were differentiated from human isogenic iPSCs expressing APOE2, APOE3, or APOE4 as previously described (Galicia Aguirre et al. 2025). Briefly, iPSCs were cultured in mTESR1 medium (STEMCELL Technology, 05850). Differentiation was initiated by inhibiting SMAD signaling with SB431542 (10 μM, Tocris, 1614) and LDN‐193189 (1 μM, Tocris, 6053) in mTESR1 medium. By day 10, the harvested aggregates were plated at low density on Matrigel‐coated (1 mL, 50 μg) 10‐cm dishes in N2B27 medium. This medium consisted of DMEM/F12 (Gibco, Thermo Fisher Scientific, 11320‐033) supplemented with 1× N2 (Thermo Fisher Scientific, 17502001), 1× B27 (Thermo Fisher Scientific, 17504001), 1× GlutaMAX (Thermo Fisher Scientific, 35050061), 1× Non‐Essential Amino Acids (Thermo Fisher Scientific, 11140050), β‐fibroblast growth factor (FGF, 25 ng/mL, PeproTech, 100‐18B), and penicillin/streptomycin (P/S, 100 U/mL, Thermo Fisher Scientific, 15140122). At day 14, neural rosettes were seeded into a Matrigel‐coated P12‐well plate in Neural Proliferation Medium. This medium contained Neurobasal (Thermo Fisher Scientific, 21103049), 1× B27 supplement (Thermo Fisher Scientific, 17504001), 1× GlutaMAX (Thermo Fisher Scientific, 35050061), leukemia inhibitory factor (10 ng/mL, PeproTech, 300‐05), and 100 U/mL P/S, supplemented with 25 ng/mL β‐FGF. NSCs derived from this process were passaged upon reaching confluency. For time‐course experiments, NSCs were plated in 96‐well plates coated with Matrigel at a density of 6.2 × 10³ cell/cm², after 2 days, cells were exposed to irradiation 5Gy and fixed using 4% PFA.

Female mice homozygous for human APOE2, APOE3, and APOE4 genes, through targeted replacement of the mouse Apoe gene, were purchased from Taconic Biosciences (Rensselaer, NY) (catalog numbers, 1547‐F or ‐M, 1548F or ‐M,1549‐F or ‐M, B6.129P2‐Apoe<tm1 (APOE*2)Mae>N9, B6.129P2‐Apoe<tm1 (APOE*3)Mae>N9, B6.129P2‐Apoe<tm1(APOE*4) Mae>N9). Mice were generated on the C57BL/6 genetic background, with the mouse APOE gene replaced with the human APOE gene under the control of murine APOE regulatory sequences. The Buck Institute for Research on Aging animal facility is an AAALAC International‐accredited institution (Unit Number 001070). All protocols and procedures described herein were approved by the Buck's Institutional Animal Care and Use Committee.

Paraffin‐embedded brain tissue sections (2 μm thickness) were deparaffinized using a standard xylene‐based protocol and rehydrated through graded ethanol. Antigen retrieval was performed in Tris–EDTA buffer (10 mM Tris base, 1 mM EDTA, pH 9.0), followed by brief rinsing in PBS. Nonspecific binding was blocked by incubating sections in blocking buffer containing PBS supplemented with 10% normal donkey serum, 5% bovine serum albumin (BSA), and 0.3% Triton X‐100 for 2 h at room temperature in a humidified chamber. Sections were then incubated overnight at 4°C with primary antibodies diluted in blocking buffer. After washing with PBS, sections were incubated with appropriate cross‐adsorbed fluorophore‐conjugated secondary antibodies diluted in blocking buffer. To reduce tissue autofluorescence, sections were treated with TrueBlack (Biotium) for 1 min, followed by three washes in PBS. Slides were mounted using ProLong Gold Antifade Mountant containing DAPI (Thermo Fisher Scientific). Images were acquired using a Zeiss LSM 980 confocal microscope. Image analysis was performed using Cellpose and Fiji (ImageJ).

Graphs and statistical analysis were performed using Graphpad Prism 10. For GABAergic neurons, we performed 3 independent experiments using unpaired Student's t‐test comparing APOE2 versus APOE4. For Ngn2 glutamatergic analysis, we performed 3 to 6 independent differentiations with technical replicates, then we performed a one‐way or two‐way ANOVA, followed by Tukey's post hoc test, which applies the family‐wise error rate (FWER). All significant values are represented by asterisks showing adjusted p values displayed in this paper as p.