In vitro characterization on the role of APOE polymorphism in human hippocampal neurogenesis

Abstract Hippocampal neurogenesis (HN) is considered an important mechanism underlying lifelong brain plasticity, and alterations in this process have been implicated in early Alzheimer's disease progression. APOE polymorphism is the most common genetic risk factor for late‐onset Alzheimer's disease where the ε4 genotype is associated with a significantly earlier disease onset compared to the neutral ε3 allele. Recently, APOE has been shown to play an important role in the regulation of HN. However, the time‐dependent impact of its polymorphism in humans remains elusive, partially due to the difficulties of studying human HN in vivo. To bridge this gap of knowledge, we used an in vitro cellular model of human HN and performed a time course characterization on isogenic induced pluripotent stem cells with different genotypes of APOE. We found that APOE itself was more highly expressed in ε4 at the stem cell stage, while the divergence of differential gene expression phenotype between ε4 and ε3 became prominent at the neuronal stage of differentiation. This divergence was not associated with the differential capacity to generate dentate gyrus granule cell‐like neurons, as its level was comparable between ε4 and ε3. Transcriptomic profiling across different stages of neurogenesis indicated a clear “maturation of functional neurons” phenotype in ε3 neural progenitors and neurons, while genes differentially expressed only in ε4 neurons suggested potential alterations in “metabolism and mitochondrial function.” Taken together, our in vitro investigation suggests that APOE ε4 allele can exert a transcriptome‐wide effect at the later stages of HN, without altering the overall level of neurogenesis per se.

Evidence suggests that the expression of notable cellular markers is highly conserved across species; for example, nestin, sex determining region Y-Box 2 (SOX2), DCX, polysialylated neuronal cell adhesion molecule (PSA-NCAM), and PROX1 are highly expressed in similar cell types across rodent, primate, and human HN (Charvet & Finlay, 2018;Miller et al., 2013). Although the persistence of HN throughout life in humans has been recently disputed (Cipriani et al., 2018;Sorrells et al., 2018), the majority of the current literature (Kempermann et al., 2018) suggests that HN is a highly conserved phenomenon, robustly observed across many mammalian species, and that newborn cells generated through HN are critical for hippocampus-dependent learning and memory (Deng et al., 2010;Gonçalves et al., 2016).
Alzheimer's disease (AD) is the most common cause of dementia affecting more than 50 million people worldwide (Prince et al., 2015). In contrast to early-onset AD, which is caused by autosomal-dominant mutations in genes such as amyloid precursor protein (APP), presenilin 1 (PSEN1), and presenilin 2 (PSEN2), late-onset AD occurs "spontaneously" with age (Van Cauwenberghe et al., 2016). An overwhelming majority of AD cases is late-onset (approximately 95%), and both early and late-onset AD share similar clinical phenotypes (i.e., severe memory loss and cognitive decline accompanied by changes in mood and behavior) (Reitz et al., 2011) and similar biological hallmarks (i.e., β-amyloid plaques, neurofibrillary tangles of hyperphosphorylated tau, and progressive neurodegeneration in multiple brain regions) (Serrano-Pozo et al., 2011).
HN is significantly reduced in post-mortem human AD brains (Ekonomou et al., 2015;Moreno-Jiménez et al., 2019;Tobin et al., 2019), and alterations in this process are one of the earliest changes observed in AD (Unger et al., 2016). HN can be a promising therapeutic target for the prevention and delay of AD onset, because it is influenced by many environmental factors that contribute to AD risk (Toda et al., 2019). However, contradictory findings on the direction and magnitude of change in HN make it difficult to pinpoint the exact approach that should be taken. Some studies show that HN is reduced in AD (Wang et al., 2004;Zhang et al., 2007), while others report that the expression of neurogenic markers is "increased" in AD (Jin, Galvan, et al., 2004;Jin, Peel, et al., 2004). One way to explain such discrepancy is to think of "increased" neurogenesis as a potential compensatory mechanism triggered to replenish the ongoing neuronal loss, while survival and maturation for the newborn neurons can ultimately result in failure (Chen et al., 2008;Sun et al., 2009). In line with this notion, transcriptomic studies on human AD brains collectively indicate that the expression of "early" neurogenic markers (i.e., markers for neural progenitor cells [NPCs] and proliferation) is "up-regulated," whereas that of "late" neurogenic markers (i.e., markers for maturation and survival) is "down-regulated" in AD (Gatt et al., 2019). Further investigations in animal studies with a higher temporal resolution would be able to clarify whether this observation in humans can be generalized to animal models. Nevertheless, a tentative summary of the existing literature can be made as follows: the overall level of HN is reduced in AD, potentially due to the failure of maturation and integration of newborn neurons, while NPCs might proliferate more over the course of disease progression as a compensatory mechanism.
The most common genetic risk factor for late-onset AD is Apolipoprotein E (APOE) polymorphism (Shen & Jia, 2016;Van Cauwenberghe et al., 2016). Two single nucleotide polymorphisms (SNPs) at the protein-coding region of APOE (exon 4) make up the following genotypes: epsilon 3, 2, and 4 (ϵ3, ϵ2, and ϵ4). The most common isoform of APOE is ϵ3, which has thymine (T) and cytosine (C) at the rs429358 and rs7412 SNP regions, respectively. In contrast, T at both regions make up the ϵ2 isoform, and C at both regions make up the ϵ4 isoform (Liu et al., 2013). It is the ϵ4 allele, which is considered the most common genetic risk factor of late-onset AD (Rebeck et al., 1993), and the frequency of ϵ4 is higher in people diagnosed with AD compared to healthy controls. The effects of ϵ4 on elevated AD risk are allele dose dependent, where the odds ratio can be increased up to 14.9 for people with homozygous ϵ4 alleles (ε2/ε4 = 2.6, ε3/ε4 = 3.2) (Farrer et al., 1997;Sando et al., 2008).
Notably, ϵ4 carriers develop AD at an earlier age (Corder et al., 1993;Farrer et al., 1997;Rebeck et al., 1993), and they also have a higher rate of progression from MCI to AD (Bonham et al., 2016).
Interestingly, postnatal neural stem/progenitor cells in the adult DG of mice express high levels of APOE  and is essential in maintaining the number of these cells throughout adulthood . Studies on injury-induced HN have shown that ϵ4 allele and knockout behave similarly, in which they can both impair dendritic arborization and reduce spine density of adult-born DGCs (Hong et al., 2016;Tensaouti et al., 2018). However, in injury-free conditions, impaired maturation of adult-born DGCs is more specific to ϵ4 and is not evident in knockout mice (Li et al., 2009). Moreover, ϵ4 hippocampal progenitors tend to proliferate more than ϵ3 progenitors, while the lack of APOE is more strongly characterized by increased gliogenesis (Li et al., 2009). Taken together, the evidence generated from various mouse studies suggests that APOE plays an important role in the overall maintenance of the hippocampal neurogenic niche, while ϵ4 allele is associated more specifically with abnormal proliferation of progenitor cells and failure of maturation in newborn neurons. Importantly, our understanding on how APOE genotype affects HN at the cellular level is yet to be obtained in humans, partially due to the difficulties of studying HN dynamics in live subjects.
While there is currently no in vitro model that specifically recapitulates "adult" human HN, evidence suggests that "embryonic" and "postnatal" HN follow a similar pattern of precursor expansion and neuroblast maturation once the radial glia-like stem cells become activated (Esp osito et al., 2005;Hochgerner et al., 2018;. Previously, Yu and colleagues developed a differentiation paradigm that recapitulates key developmental events that occur in both embryonic and postnatal HN, demonstrating that human pluripotent stem cells can be differentiated into hippocampal DGClike cells expressing PROX1 using this protocol (Yu et al., 2014). In this study, we used a similar in vitro model of HN based on Yu and colleagues' method that can generate DGC-like cells from human induced pluripotent stem cells (iPSCs). We aimed to characterize the phenotypes of HN according to various APOE genotypes at the cellular level using this model and isogenic human iPSCs.

| Cell lines
Isogenic iPSCs (male origin) were obtained from the European Bank for induced pluripotent Stem Cells (EBiSC). Full data on reprogramming and characterization of BIONi010-C-2 (ϵ3 genotype, denoted E3 in this article; RRID:CVCL_II81), BIONi010-C-4 (ϵ4 genotype, denoted E4 in this article; RRID:CVCL_II83), BIONi010-C-6 (ϵ2 genotype, denoted E2 in this article; RRID:CVCL_II85), and BIONi010-C-3 (APOE knockout, denoted KO in this article; RRID:CVCL_II82) lines can be found in the report by Schmid and colleagues . A corrigendum to this article shows that E3, E4, and E2 lines have only one functional ϵ3, ϵ4, ϵ2 allele of APOE expressed with the correct genotype, respectively; the KO line does not express any functional APOE (Schmid et al., 2020). 14170-161) and incubated with 1 ml/well room temperature Versene for 3-4 min. After aspirating Versene from each well, cells were detached gently (2-3 wells at a time) with 5-6 ml of room temperature StemFlex™, using 5 ml or 10 ml stripettes. Detached cells were collected in new 50 ml conical tube(s) and were gently broken down with 5 ml or 10 ml stripettes until the size of cell clumps were reduced to that of "dots" when viewed with unaided eyes. Passaging ratio was kept between 1:6 and 1:18 depending on the experimental requirements. Spontaneously differentiated iPSC colonies were regularly cleaned prior to passaging with sterile aspirator pipettes with 10/20 μl pipette tips inserted at the end. Time spent outside the incubator for cleaning was always kept under 5 min.

| Replating iPSCs
All stages of directed differentiation were performed in sterile six-well NUNC™ plates except for terminal plating for DGC-like neuronal differentiation. A schematic diagram summarizing the procedure is shown in Figure 1 (top panel). To replate iPSC colonies for directed differentiation, cells were first maintained to reach 70%-80% confluence and then lifted with 1 ml/well room temperature Versene. After aspirating the Versene, 1 ml/well room temperature StemFlex™ was introduced to all wells, and iPSC colonies were gently but swiftly scraped off with sterile cell scrapers (wedge facing the cells). Lifted colonies were then collected in new 50 ml conical tube(s) with either 5 ml or 10 ml stripettes. The colonies were gently pipetted up and down with a P1000 until they were broken down to small granules that were still visible to unaided eyes. Cells were then transferred to new Geltrex™ plates at a passaging ratio of 3:2, so that the confluence could reach near 100% in 24-48 hReplated iPSCs were incubated at 37 C, 5% CO 2 , 5% O 2 conditions, and if confluence was not near 100% in 24 h, medium was changed to fresh StemFlex™ (3 ml/well) and incubated for an additional 24 h. Replated iPSCs that failed to reach near 100% confluence after 48 h were discarded and did not proceed to directed differentiation.

| Initiation of directed differentiation
Directed differentiation of replated iPSCs was initiated (Day 0) by changing the following culture conditions. StemFlex™ was replaced with N2B27 + 4i medium. This was a 1:1 mixture of N-2 medium

| NPC expansion
On Day 7 of directed differentiation, the first neural passaging was done at 1:1 ratio. Cells were washed with 1 ml/well room temperature HBSS and incubated with 1 ml/well 4 C Accutase (Thermo Fisher, Cat# A11105-01) for 3-4 min. Dissociated cells were collected into new 15 ml conical tube(s) that already contained room temperature DMEM/F12 at twice the volume of Accutase used. Importantly, the collection of cells in Accutase into conical tube(s) was done with no more than five times of pipetting with a P1000 to ensure passaging of cells in "small clumps" rather than "single cells." Remaining clumps of cells in the plate that were not lifted at the first attempt were all collected with 1 ml/well room temperature DMEM/F12 using a P1000, and they were pooled together with the cells collected at the first attempt. Cells were centrifuged at 900 revolutions per minute (RPM) for 2 min twice. Prior to the second centrifugation, cells were resus-

| Terminal plating for DGC-like neuronal differentiation
For DGC-like neuronal differentiation, NPCs expanded as described above were terminally plated on either sterile 6-well or 96-well NUNC™ plates coated with poly-L-ornithine (Sigma Aldrich, Cat# P3655) and laminin (Sigma Aldrich, Cat# L2020) (POL) (Day 20/21). POL plates were prepared by incubating each well with 100 μg/ml poly-ornithine for 3 h at 37 C (1.5 ml/well for 6-well plates and 50 μl/well for 96-well plates) followed by 20 μg/ml laminin overnight at 37 C (2 ml/well for 6-well plates and 75 μl/well for 96-well plates). Terminal plating procedure was identical to the third and final neural passaging, except for the passaging ratio, which was at 1:6 (or 20,000 cells/cm 2 ).
On the next day of terminal plating (Day 21/

| Gene expression analysis
, and (f) NEUROD1 were normalized to GAPDH expression. Day 0 of E3 was used as the reference sample for each gene. Two-way ANOVA with Bonferroni correction. n = 3. Mean with SD shown. Adjusted p values: 0 < *** < 0.001 < ** < 0.01 < * < 0.05, when each cell line was compared to E3 on each day of directed differentiation. ing the border of each image were eliminated to ensure that a given nucleus is not counted more than once for downstream analysis ("Select Population" module). After calculating the "Area" and "Roundness" features of nuclei ("Calculate Morphology Properties" module),

| Immunocytochemistry
further filtering was applied based on these calculated features so that viable and nonclumped nuclei were chosen for downstream analysis ("Select Population" module). On this filtered population of nuclei, cytoplasmic region surrounding the nucleus for each cell was marked as a ring-shaped region extending away from the nucleus ("Select Region" module, "Resize Region" method). The intensity features of this cytoplasmic region were calculated on images acquired from Alexa 488 and Alexa 568 channels ("Calculate Intensity Properties" module). Then, cells that met the threshold criteria for each channel intensity features were marked as being "positive" for a given cellular marker ("Select Population" module). The percentage of positive cells were reported at the end of analysis ("Define Results" module). Multiple negative-control images were used (cells incubated only with secondary antibodies) to configure the correct threshold settings for each channel intensity. An example analysis pipeline is shown in Figure S1.  (Overall et al., 2012). DEGs of each timepoint (i.e., gene list) was matched against the "gene sets" curated in MANGO, which contains information on which "cellular stage" of HN a given gene expression is found to be "present" or "absent" in vivo (filtered by "Process/ Outcome = Expression" AND "Effect = (+) OR (À)"). The DEG lists were sorted by log2 fold change values prior to analysis, and if a gene in the MANGO gene set (e.g., "Granule cell neuron") was also found in the DEG list, the "running enrichment score" for that gene set was increased in proportion to the log2 fold change value of the gene. The results were visualized using the gseaplot2() function of the enrichplot package (Wu et al., 2021).

| Exploratory analysis on transcriptomic data shows little evidence of APOE genotype-dependent effect
The qPCR and ICC data described thus far suggest the following: (1) despite significant differences at certain stages of differentiation,  Figure 9a). In line with this observation, the top 35 genes that varied the most across all samples distinguished the different "timepoints" of differentiation most clearly (Figure 9b).
The Z-scores derived from the normalized expression matrix provided the following updates to our qPCR characterization. First, DCX expression was the highest at Day 43, but contrary to the qPCR data, the levels were comparable between E3 and E4. Second, FOXG1 expression was confirmed to be the highest at Day 18, but contrary to the qPCR data, E3 and E4 had similar levels of expression at this timepoint, and it was Day 43 when the expression was down-regulated in E4 compared to E3. Finally, APOE expression was confirmed to decrease with differentiation, and its level is also relatively higher in E4 at Day 3.
Exploratory analysis of the top 35 most variable genes also revealed changes in the following genes that were not directly charac- NPCs and neural stem cells (Andersen et al., 2014), was found to decrease in E3 and not as much in E4 at Day 43. Finally, a similar protracted expression was observed for HES5, another pro-neural transcription factor highly expressed in neurogenic neural stem cells (Jin et al., 2003), in which the clear decrease from Days 18 to 43 was evident in E3 but not in E4.

| Gene set enrichment analysis on differentially expressed genes confirm similar cellular identity for E3 and E4
To compare the cell fates assumed by E3 and E4 cells during directed differentiation, we conducted a gene set enrichment analysis (GSEA) Genes with either "present (+)" or "absent (À)" effect on the outcome "expression" according to the MANGO database were used for enrichment analysis. Red box indicates the top two enriched gene sets in both E3 and E4 cells: "Determined progenitor (Type 2b)" and "Neuroblast-like cell (Type 3)." 39 genes were differentially expressed at Days 3, 18, and 43, respectively ( Figure 15). Notably, one gene that was consistently found to be down-regulated in E4 across all stages was coiled-coil-helix-coiledcoil-helix domain containing 2 (CHCHD2). We also noted that the  (Lee et al., 2020), and we report here a similar finding in our model of HN. These observations collectively suggest a role of APOE in stem cell maintenance/differentiation, in line with other independent studies on mice (GEO accession: GSE22908) (Polo et al., 2010), and humans (GEO accession: GSE20750) Saito et al., 2011;Tateno et al., 2011), where APOE was consistently found to be expressed more highly in "less" differentiated cells. With regards to genotype-dependent differences, here we show that APOE is more highly expressed in E4 and E2 compared to E3 at the iPSC stage, prior to directed differentiation, although the exact "functional significance" of these differences continues to remain an avenue for further research. Genes with either "present (+)" or "absent (À)" effect on the outcome "expression" according to the MANGO database were used for enrichment analysis. Only two gene sets were identified to be enriched for E4 cells: "Doublecortin immunoreactive cell" and "Granule cell neuron." The top enriched gene set in both E3 and E4 is "Granule cell neuron." expression was significantly lower than E3 at many timepoints, had from E3 are likely to be related to alterations in metabolism and mitochondrial function. CHCHD2 has been previously shown to localize at the intermembrane space of mitochondria (Aras et al., 2015;Funayama et al., 2015), and decreased expression and/or loss has been associated with mitochondrial dysfunction (Funayama et al., 2017). Long-term damage in (mitochondrial) metabolism can detrimentally affect brain regions like the DG, where energy demands are high due to ongoing neurogenesis, even though the cells are equipped with the correct molecular machinery to generate DGCs.
Interestingly, altered metabolism linked to changes in brain function is one of the most robust phenomena reported consistently in various observational studies of ε4 carriers (Liu et al., 2015). However, it should be noted that this study did not examine the functional differences of E4 and E3 cells, and further experiments would be necessary to validate the phenotypic inferences one can make from our transcriptomic dataset. Therefore, we propose that follow-up studies should focus on elucidating the relationship between altered (mitochondrial) metabolism and APOE genotype in the context of HN.
In addition, we recognize the following limitations of this study. The isogenic cell lines contained only one functional allele of APOE despite expressing the correct isoform (Schmid et al., 2020) (except KO, which had no functional APOE expressed altogether).
Although the expression of both alleles in each cell line would have either provided "more" or "better" insight into the differences between APOE genotypes, our data alone suggest that SNPs on a "single" allele can exert transcriptome-wide effects that clearly resonate with other independent studies that found "altered metabolism and mitochondrial function" as a potential mechanism of E4 under homozygous conditions. Another limitation is that our model does not distinguish embryonic and adult HN. The existing literature on APOE genotype and HN hints at adult neural stem cell pool exhaustion  and maturation failure of adultborn neurons (Li et al., 2009) in the DG as important phenotypes in rodents expressing either no APOE or human ε4, respectively.
Both neural stem cells and newborn neurons in the "adult" DG have been shown to bear significant functional differences from their embryonic counterparts (Cole et al., 2020;. Therefore, a more specific modeling of "adult" HN could be yield further insights that would have been difficult to obtain using our model. Furthermore, it is important to note that in vitro models of HN, such as the one used in this study, cannot fully recapitulate the complexity of the in vivo DG niche. The DG is known to be highly vascularized (Shen et al., 2019) and consists of many types of cells, such as inhibitory neurons, glial cells, oligodendrocytes, and endothelial cells, all of which have a significant influence on the course and outcomes of HN (Li et al., 2009;Morrens et al., 2012;Seki, 2003). A more "in vivo-like" model of HN that takes many of these cell types into account would be especially relevant for the investigation of APOE, since it is known to be highly expressed in glial cells such as astrocytes and microglia (Liu et al., 2013). Nevertheless, the simplicity of the model allowed us to isolate the "cell-autonomous" effects of APOE in HN, and we show here that APOE expression in neural stem cells, in and of itself, is crucial for HN.
In summary, our in vitro model of human HN and isogenic APOE lines demonstrate that APOE is differentially expressed at the stem cell stage, while the phenotypic divergence of E4 from E3 became more prominent at the neuronal stage of differentiation, without substantial changes to DGC-like neuronal differentiation. Genes associated with "maturation of functional neurons" were more clearly expressed in E3 neurons compared to E4, while genes uniquely expressed in E4 neurons indicated alterations in metabolism and mitochondrial function. Future investigations on these potential "vulnerabilities" could provide better insight into the mechanistic link between defects in metabolism and hippocampal function, both of which are frequently observed in ε4 carriers.

CONFLICT OF INTEREST STATEMENT
The authors declare no competing interest.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are openly available at Open Science Framework (osf.io/w67cd).