Flexible and Accurate Substrate Processing with Distinct Presenilin/γ-Secretases in Human Cortical Neurons

Visual Abstract


Introduction
Alzheimer's disease (AD) is the most common neurodegenerative dementia and is characterized by specific neuropathological lesions, including senile plaques, in the brain parenchyma of afflicted patients. The senile plaques are mainly composed of b -amyloid (Ab ) peptides and appear several decades before the onset of clinical symptoms, leading to the widely accepted amyloid hypothesis (Hardy and Selkoe, 2002). A few hundred mutations in the amyloid b precursor protein (APP) and presenilin (PSEN1 and PSEN2, hereafter referred to as PS1 and PS2) genes have been identified in familial cases of AD (FAD) to date (http://www.alzforum.org/mutations). Notably, most cases of FAD are attributed to mutations in the PS1 and PS2 genes, highlighting the importance of PS mutations in AD pathogenesis. PS1 and PS2 are expressed throughout life  and serve as an integral catalytic subunit of the g -secretase complex (Kimberly et al., 2003;Takasugi et al., 2003). g -Secretase generates Ab by a stepwise processing of membrane-tethered APP C-terminal fragments (APP-CTFs), which are the initial ectodomain-shed products of APP by b -secretase (De Strooper et al., 1998;Vassar et al., 1999), and most FAD-linked mutations cause a relative increase in highly toxic longer Ab species, such as Ab 42 (Borchelt et al., 1996).
Because of the presence of homologs of the PS and Aph-1 genes in vertebrates, g -secretase exhibits heterogeneity depending on which homolog is assembled into the complex. PS has two homologs, PS1 and PS2, with ;67% sequence homology (Levy-Lahad et al., 1995;Rogaev et al., 1995;Sherrington et al., 1995) and overlapping and separate functions (De Strooper et al., 2012). In particular, PS1 is more important during development: PS1 germline knock-out (KO) mice die perinatally, whereas PS2 KO mice have few detectable phenotypes Wong et al., 1997;Steiner et al., 1999). However, mice lacking both PS1 and PS2 die much earlier than PS1 KO mice (Donoviel et al., 1999), and the neurogenesis defects in PS1/PS2-null neural progenitor cells (NPCs) are much more severe than those in PS1-deficient NPCs (Handler et al., 2000;Hitoshi et al., 2002;Kim and Shen, 2008). In terms of Ab generation, PS1 and PS2 exhibit different properties in nonneuronal cells and/or nonhuman experimental systems, most of which express exogenous PS, and PS2/g -secretase shows less total proteolytic activity than PS1/g -secretase (Lai et al., 2003;Yonemura et al., 2011;Pintchovski et al., 2013). However, whether distinct PS/g -secretase complexes show similar enzymatic activity in terms of substrate processing in human neurons remains unclear thus far.
Recent advancement in the induced pluripotent stem cell (iPSC) technique made it possible to access human neural cells for molecular and cellular research of neurologic disorders (Takahashi et al., 2007). Since then, substantial studies have reported many pathologic AD phenotypes using iPSC-derived neural cells from AD patients, including FAD (Penney et al., 2020). Mutations in the PS gene expectedly led to a significant increase in the Ab 42/40 ratio and, in some cases, tau pathology in iPSCderived neurons (Yagi et al., 2011;Woodruff et al., 2013;Imaizumi et al., 2015;Kondo et al., 2017;Ishikawa et al., 2020;Sho et al., 2020). Despite extensive investigations, however, PS physiological functions from the aspect of PS/g -secretase heterogeneity remain to be investigated in human neural cells.
In this study, to investigate distinct PS/g -secretase complexes in human cortical neurons, we generated PS1 conditional KO (cKO) iPSC lines with or without additional PS2 KO alleles using the CRISPR/Cas9 system. We clearly demonstrated the substrate specificity between PS1/g -secretase and PS2/g -secretase; N-cadherin is cleaved solely by PS1, while APP/Notch is targeted by both PS1 and PS2. Moreover, using a specific antibody against the active g -secretase complex, we showed the differences in the endogenous subcellular localization between PS1/g -secretase and PS2/g -secretase in human neurons for the first time. Together, these results provide direct experimental evidence showing the intrinsic heterogeneity of PS/g -secretase in human neurons and promising insights into the molecular mechanism of PS/ g -secretase dysfunction in AD pathogenesis.

Neuronal induction
Cortical neuronal induction of iPSCs was performed as described previously with some modifications (Telezhkin et al., 2016;Sato et al., 2021), which will be published elsewhere (the patent publication number, WO/2020/ 045578). Briefly, semiconfluent feeder-free iPSCs were cultured for 14 d in neural medium with dual SMAD inhibitors and Wnt inhibitor. The consequent NPCs were dissociated and seeded at a density of 5 Â 10 4 cells/cm 2 on multiwell plate coated with poly-ornithine and Matrigel (Corning). Terminal differentiation was induced in neural medium supplemented with B27 (Invitrogen) and 10 mM DAPT (Sigma) for 6 d. After day 6, the culture was maintained in BrainPhys Neuronal Medium (Stemcell Technologies) supplemented with 10 ng/ml BDNF, 10 ng/ml GDNF, 200 mM ascorbic acid, 0.5 mM dbcAMP, and changed medium every 3-4 d with a half volume until day 45.

Generation of conditional PS1 KO and PS2 KO iPSCs
To construct a targeting vector for the PS1 allele, a 4.41-kb genomic DNA, which contains the exons 2-3, was first amplified by PCR using genomic DNA of 201B7 iPSC as a template, with a primer pair 59-CCTGGCCTC AAGTAGTAACACCCAT-39 and 59-CACAGCAGCCCACA AAAGGAAAACT-39, and subcloned into pCR-BluntII TOPO (Invitrogen). After sequence confirmation, a 1.18-kb NdeI-NotI DNA fragment (L1), a 1.12-kb NotI-SbfI fragment (L2) encompassing the second and third exon of the PS1 gene, and a 1.08-kb BamHI/SacII-BamHI fragment (R1) were amplified by PCR using the above genomic fragment as a template. The PCR primers are 59-TACCATATGAGTCTCACTC TGTTGCCCAGG-39 and 59-ATTGCGGCCGCAGGACGG ACAGATAACATG-39 for the left arm L1 (the underlined sequences are for NdeI and NotI), 59-TAAGCGGCCGCGG GCATTGTGATAAGG-39 and 59-ATGCCTGCAGGAACCCT TAGAACTTCTACAC-39 for the left arm L2 (the underlined sequences are for NotI/SacII and SbfI), and 59-ATTGGATC CGCGGAGGGATTCAGGAAAAGAAC-39 and 59-ATTGGAT CCTGGGCACATCAAAACTTCC-39 for the right arm R1 (the underlined sequences are for BamHI/SacII and BamHI). These genomic fragments were used as 59 (L1 and L2) and 39 (R1) homologous regions in the PS1 targeting vector. Both L1 and L2 fragments were subcloned into NdeI-SbfI sites in the pUC-FRT-PGK-neo-FRT plasmid (#519), in which PGK-neoDtk (a fusion protein of a neomycin-phosphotransferase and a truncated version of herpes simplex virus type 1 thymidine kinase) cassette was flanked by two FLP recognition target (FRT) sites to allow removal of the neoDtk gene by FLP, as its presence can suppress transcription of the target gene. After sequence confirmation of homologous arm region, a loxP fragment was inserted into a SacII site in the left arm, which was a vicinity of the junction between L1 and L2 fragments. A loxP fragment was generated by annealing the following oligonucleotides: 59-ATCGATATAACTTCGTATAGCATACATTATACGAAGTTAT-TTGC-39 and 59-AAATAACTTCGTATAATGTATGCTATAC GAAGTTATATCGATGC-39 (a 34-nt loxP sequence is underlined). Next, a right arm fragment (R1) was subcloned into BamHI site of the left-arm plasmid and confirmed by sequencing. A loxP fragment was also inserted into a SacII site in the right arm to generate the PS1 targeting vector (neo). PGK-neoDtk cassette was replaced with PGK-puDtk (a fusion protein of a puromycin N-acetyltransferase and a truncated version of herpes simplex virus type 1 thymidine kinase) by Gateway BP reaction to generate the PS1 targeting vector (puro). To generate a plasmid expressing single guide RNA (sgRNA) for PS1 genome editing, the following oligonucleotides were annealed and inserted into a BbsI site of pSpCas9(BB)-2A-GFP (PX458), which can simultaneously express Streptococcus pyogenes Cas9 (SpCas9) and GFP: 59-CACCGTTATCTGTCCGTCCTGCCTT-39 and 59-AAACAAGGCAGGACGGACAGATAAC-39 for PS1 sgRNA1, and 59-CACCGTAGAAGTTCTAAGGGTTCAA-39 and 59-AAACTTGAACCCTTAGAACTTCTAC-39 for PS1 sgRNA2 (the targeted PS1 sequences are underlined).
To generate iPSCs carrying the targeted/floxed PS1 alleles, 201B7 iPSCs were electroporated with a targeting vector (puro) along with two sgRNA-expressing plasmids. Puromycin was applied to the culture at 0.5 mg/ml 24 h later and the surviving iPSC clones were picked following 7 d of puromycin selection. Genomic DNAs around the targeting region of total 12 iPSC clones were first amplified by PCR using the primer pair for PS1 genomic cloning as shown above, which encompasses both homologous recombination arms. Next, nested PCR was performed using the primer pair 59-CTCCTGGCTGAGTCTGCGAT-39 and 59-AGAACCGCCTGAGACACCAA-39, which encompasses the upstream loxP sequence. Two iPSC clones were positive for proper homologous recombination in one allele, giving rise to 345-and 392-bp bands, which represent the wild type (WT) and the targeted allele, respectively. This heterozygous clone was further electroporated with a targeting vector (neo) along with two sgRNA-expressing plasmids. G418 was applied to the culture at 100 mg/ml 24 h later and the surviving iPSC clones were picked following 10 d of G418 selection and screened as previously. Two out of 30 iPSC clones were confirmed to carry the correct homologous recombination at both alleles. We verified proper recombination by sequencing to confirm the presence of the single loxP sites in targeted allele. To generate iPSCs bearing floxed and deleted PS1 alleles, these homozygous targeted PS1 clones (#245 and #249) were transfected with pCAGGS-FLP and pCAGGS-Cre, respectively, and selected under 10 mM ganciclovir treatment for 10 d. In this study, most experiments were conducted using one targeted PS1 clone (#245)-derived cells.

Colony-forming assay
Colony-forming assay using iPSCs was performed as described previously (Imaizumi et al., 2015;Fujimori et al., 2017). Briefly, iPSCs were pretreated for 6 d with 3 mM SB431542 (Tocris) and 150 nM LDN193189 (StemRD). To make clonal neurospheres, the cells were then dissociated and seeded at a density of 10 cell/ml in media hormone mix (MHM) with selected growth factors and inhibitors under hypoxic conditions. The growth factors and inhibitors included 20 ng/ml FGF-2, B27 supplement (Invitrogen), 2 mM SB431542 and 10 mM Y-27632 (Calbiochem). We infected lentiviruses expressing either mCherry-nls-Cre or mCherry-nls-DCre at multiplicity of infection (MOI) %1 just at the beginning of the culture. Here, we defined the day on which neurosphere culture was started as day 0. On day 12, primary neurospheres were then dissociated and seeded at a density of 100 cell/ml to make secondary neurospheres. For flow cytometry analysis, dissociated neurospheres were suspended in PBS supplemented with 4% fetal bovine serum (FBS) and analyzed mCherry fluorescence with a FACSAria (BD Biosciences) using an 85-mm nozzle, to calculate the lentivirus infected population. On day 18, secondary neurospheres were then dissociated and the fluorescence of the cells was analyzed by flow cytometry.

Quantitative reverse transcription-PCR (qRT-PCR)
RNA extraction and qRT-PCR were performed as described previously (Imaizumi et al., 2015;Fujimori et al., 2017). Briefly, total RNA was purified with RNeasy Micro kit or RNeasy Mini kit (QIAGEN) and reverse-transcribed in the presence of random hexamers. qPCR reactions were performed using TB Green Premix Ex Taq II (TAKARA BIO Inc.) in ViiA 7 System (Applied Biosystems) with cDNA and gene-specific primers. Analysis was performed with at least three independent cultures and threshold cycle (Ct) values of interest were normalized to GAPDH. The primer pairs used in this study are as fol-
Production of recombinant lentiviruses is achieved by transfecting HEK293T cells with three plasmids. Lentiviruses were harvested 48 h after transfection by collecting the medium from transfected cells, and filtrated with a 0.45-mm filter. Titer of the lentivirus was estimated by measuring the EGFPpositive or mCherry-positive cells with fluorescent microscopy, following the infection of diluted lentivirus to HEK293T cells. Neurons were infected with each lentivirus at ;3 of MOI.

PS DKO MEF cultures and Ab production assay
Stable mouse embryonic fibroblasts that lack both PS1 and PS2 and express human APP gene carrying Swedish mutation heterologously (DKONL; Herreman et al., 2000;Watanabe et al., 2005) were maintained in high glucose DMEM (Invitrogen) supplemented with 10% FBS and 0.1 mg/ml puromycin. DKONL cells were infected with PSexpressing lentivirus and the culture medium was changed 1 d after lentivirus infection. The culture medium was collected following 48 h incubation and spun down for removing cell debris and stored at À80°C until use for Ab ELISA.
For immunofluorescent analysis, 96-well plates were imaged with an IN Cell Analyzer 6000 high-content cellular analysis system (GE Healthcare). A set of 3 Â 3 fields was collected from each well using the 20Â objective lens. Comprehensive analysis was performed with IN Cell Developer Toolbox version 1.9 (GE Healthcare). First, the lentivirus (DCre or Cre)-infected neuronal population was identified by signals for mCherry and MAP2 fluorescence, which were defined as mCherry 1 nuclei that were .50 mm 2 and with intensity levels that were appropriate brightness of intact cells. In each traced MAP21 region, puncta positive for glycosylated nicastrin and LAMP1 (late endosomes/lysosomes) or EEA1 (early endosomes) were detected; from these images, the fluorescence area of each puncta and their colocalization was measured and analyzed.

ELISAs for Abs
iPSC-derived neurons were differentiated by plating almost the same number of NPCs and cultures were maintained in 48-well plate until analysis (day 45). Medium was fully changed with 500 ml/well of fresh medium 48 h before the harvest. The collected medium was centrifuged to remove insoluble material and stored at À80°C until analysis. The remaining neuronal cells were lysed in RIPA buffer and protein concentration was measured by BCA Protein assay (Pierce). Ab 40 and Ab 42 levels in the conditioned medium were measured using commercial kits, Human b Amyloid (1-40) ELISA kit II (catalog #298-64601) and Human b Amyloid (1-42) ELISA Kit High Sensitive (catalog #296-64401) from Wako, respectively, according to the manufacturer's protocol. Each Ab concentration was normalized by protein levels of the culture.

Statistical analysis
Data are presented as the mean value 6 SEM, except Figures 3C,D, 4D (mean 6 SD). The data in Figures 1-7, Extended Data Figures 5-1, 7-1 were analyzed using oneway analysis of variance and post hoc Dunnett's or Tukey's test. The data in Figure 4C and Extended Data Figure 5-1A were analyzed using Student's t test. Statistical significance was defined as p , 0.05. The data were analyzed using R version 4.0.0 (The R Foundation).

Ethics approval and consent to participate
Human ethics approval was obtained from the Ethics Committee in Keio University School of Medicine (approval number 20080016).

Results
Expression profile of each c-secretase subunit in human iPSC-derived cortical neurons g -Secretase is widely expressed in nearly all cell types of multicellular organisms. However, the expression of every subunit of g -secretase in human neural development remains to be determined, whereas many studies have focused Ab production in human iPSC-derived neurons (Yagi et al., 2011;Yahata et al., 2011;Israel et al., 2012;Sho et al., 2020). We first examined how g -secretase subunits are expressed throughout human neural differentiation using the efficient and robust differentiation protocol of iPSC-derived cortical neurons (Fig. 1A). The human neurons derived from a healthy control iPSC line exhibited characteristics of mature forebrain cortical neurons with b III-tubulin-positive intricate neurites, and over 80% of them were positive for NeuN and TBR1 at day 45 (Fig. 1B). Next, to investigate the expression of each g -secretase subunit during neural differentiation, we performed qRT-PCR using iPSCs, NPCs at day 14 and welldeveloped cortical neurons at day 45 (Fig. 1C). Both catalytic homologs, PS1 and PS2, were gradually expressed in the course of differentiation and showed the highest expression in cortical neurons. Intriguingly, PS2 mRNAs were drastically upregulated in neurons by ;4-fold compared with those in iPSCs and NPCs, suggesting that PS2 plays important roles in mature human neurons. Among other essential subunits, Aph-1a and Pen-2 were specifically increased in cortical neurons, whereas nicastrin and Aph-1b were expressed at comparable levels during differentiation. APP mRNAs were also upregulated in cortical neurons compared with those in iPSCs and NPCs, indicating that APP function is more important in mature neurons such as a putative neuronal adhesive protein (Sosa et al., 2017). More strikingly, expression of BACE1, a critical and rate-limiting enzyme for Ab production, was significantly increased by ;20-fold in neurons compared with that in iPSCs. These results suggest that iPSC-derived mature neurons are the most important source of Ab rather than iPSCs and NPCs.
Generation of human stem cell models for assessing c-secretase physiological functions g -Secretase is essential for embryonic development exclusively via regulation of the Notch pathway (De Strooper et al., 1999). Indeed, the deficiency of crucial components of g -secretase such as the PS1, Aph-1a, or Pen-2 genes causes embryonic or perinatal lethality in mice Wong et al., 1997;Ma et al., 2005;Bammens et al., 2011), which phenocopies Notch1-null mice (Swiatek et al., 1994). As the Notch pathway is important for the generation and maintenance of NPCs (Handler et al., 2000;Kim and Shen, 2008), simple KO of PS1 could cause a deleterious disturbance in neural differentiation from PSCs. We thus used a cKO method using a Cre/loxP system to uncover the normal functions of g -secretase in iPSC-derived neural lineages. More importantly, this state-of-the-art strategy can circumvent the frequent problems caused by iPSC clonal variability as well (Cahan and Daley, 2013;Liang and Zhang, 2013).
The PS1 targeting vector includes the 59 homologous region (;2.3 kb) containing exons 2-3, a selection cassette (PGK-puroDtk or PGK-neoDtk) flanked by FRT sequences and the 39 homologous region (;1.1 kb), in which exons 2-3 are adjacent to two loxP sites in the 59and 39-homologous regions ( Fig. 2A). Homozygous targeted PS1 iPSCs, in which the proper homologous recombination events occurred successfully, were confirmed by sequencing. Because PS1 is expressed in iPSCs, we first examined genome-edited iPSCs to determine whether the floxed exons can be recombined and whether expression of the PS1 protein is actually abrogated. Genomic DNAs from homozygous targeted PS1 iPSCs transfected with a plasmid expressing FLP or Cre were analyzed with a primer pair encompassing the loxP sites and the selection cassette. FLP and Cre can delete a region flanked by FRT and loxP sequences, respectively (Fig. 2B). We next performed Western blotting to investigate whether removal of exons 2-3 leads to ablation of PS1 proteins. PS1-NTFs and PS1-CTFs were almost completely eliminated in the iPSCs carrying homozygous deleted alleles, whereas the expression levels were comparable among parental cell (201B7) and iPSCs harboring the targeted or floxed alleles (Fig. 2C). These results clearly demonstrated that this PS1 cKO system works correctly on Cre introduction, by which PS1 proteins are eliminated. In the following experiments, we used iPSCs harboring homozygous floxed PS1 alleles, where removal of the selection cassette from targeted alleles occurred successfully.
Deleterious effects of presenilin deficiency on Notchdependent NPC maintenance in PS-deficient iPSCderived neural stem cells Loss of PS1 or PS1/PS2 results in the depletion of neural stem cells in mice, as the Notch signaling pathway is Figure 2. Generation of PS cKO iPSCs. A, Targeting strategy for PS1 cKO. In the targeting vector, exons 2/3 are flanked by a loxP site (red arrowhead) and a PGK-neo (or PGK-puro) selection cassette followed by a loxP. The neo selection cassette is flanked by two FRT (FLP recognition target) sites (black circles), so that the cassette can be removed by FLP recombinase. Green lines were the genomic sequences of homologous regions. iPSC clones were screened by genomic PCR using the primer pairs (arrows) to discriminate iPSC cells carrying the proper recombination from untargeted ones. The homozygous targeted clones were transfected with a plasmid expressing Flp gene to generate homozygous floxed PS1 iPSC. In the presence of Cre, the floxed allele will be recombined to produce the deleted PS1 allele. B, Genomic PCR for respective PS1 alleles. Genomic PCR was performed using primer pair encompassing two loxP sites, following transfection with plasmid expressing Flp or Cre recombinase. Parental iPSC line (201B7) and transfectant with empty vector (nc) were also shown. C, PS1 protein expression in iPSC clone carrying respective PS1 allele. Quantitative analysis shows nearly eliminated levels of PS1 proteins in total lysates isolated from the iPSC clone carrying deleted allele after Cre transduction, compared with other iPSC clones, using rabbit polyclonal antibodies specific for PS1-NTF and PS1-CTF. Data represent the mean 6 SEM (n = 3 of independent culture batches). ns, not significant; ***p , 0.005 by Tukey's test among the genotypes. D, KO strategy for PS2 gene. sgRNAs were transfected along with Cas9 proteins, to delete the genomic region around the exon 5. E, Sequence of both alleles around the PS2 targeted region from clone #12 (nearly negative for PS2 protein was confirmed by PS2 Western blot analysis of single clones transfected with PS2 sgRNA and Cas9 proteins as shown in Extended Data Fig. 2-1). One allele was completely deleted (lower), while another allele was inverted between two sgRNA sites (upper). severely compromised (Handler et al., 2000;Hitoshi et al., 2002;Kim and Shen, 2008). To examine the effect of PS deficiency on Notch signaling, we first induced the differentiation of fPS1/fPS1;PS2 1/1 (WT), PS1 À/À ;PS2 1/1 (DPS1), where the PS1 gene was already KO out by Cre, and fPS1/fPS1;PS2 À/À (DPS2) iPSCs into NPCs at day 14 and performed qRT-PCR to examine several genes related to the Notch pathway (Fig. 3A). OCT4 mRNAs were nearly absent in NPCs, whereas expression of PAX6, a forebrain NPC marker, was induced, suggesting a successful escape from the pluripotency state. As expected, PS1 and PS2 mRNAs were almost completely eliminated in iPSCs/NPCs derived from DPS1 and DPS2 iPSCs, respectively. Both NOTCH1 and NOTCH2 expression were increased significantly in NPCs compared with those in iPSCs without any genotypic effects. Interestingly, the Notch target genes HES1 and HES5 were also robustly upregulated in NPCs, especially HES5, which was upregulated by ;5000-fold, indicating that in iPSC-derived human NPCs, HES5 is a main downstream effector of Notch signaling. Surprisingly, the expression of HES1 and HES5 was slightly but not significantly decreased in DPS1 NPCs compared with those in WT and DPS2 NPCs (p = 0.225 for HES1 and p = 0.185 for HES5 compared with the WT by Dunnett's test). The inefficacious effect of PS1 deficiency on HES5 reduction in human NPCs is inconsistent with the results in prior literature showing a significant decrease in Hes5 in the telencephalon and anterior diencephalon in PS1-null mouse (Handler et al., 2000). These results suggest that PS2/g -secretase-mediated Notch activity can sufficiently compensate for the lack of PS1 in human NPCs.
As the loss of PS1 alone in NPCs tended to affect Notch signaling negatively, albeit not significantly, we further investigated the effects of PS deficiency on Notchmediated neural stem cell potency. To avoid any possible disadvantages for iPSCs caused by the long-term absence of PS, we applied a clonal colony-forming assay using fPS1/fPS1;PS2 1/1 and fPS1/fPS1;PS2 À/À iPSCs, by which PSCs can be initially induced in serum-free medium to generate neurospheres (Chaddah et al., 2012;Imaizumi et al., 2015Imaizumi et al., , 2018Fujimori et al., 2017). To investigate whether a lack of PS causes any effects in iPSCderived neurospheres, lentivirus expressing mCherry-Cre or mCherry-DCre was used for infection at a MOI %1 at the very beginning of the first neurosphere formation (Fig. 3B). We then measured the mCherry 1 cell population by flow cytometry following dissociation at the end of the first and second neurosphere formation (Fig. 3C,D; Extended Data Fig. 3-1). Intriguingly, mCherry 1 cells in the second Figure 3. Maintenance deficit in iPSC-derived NPCs lacking PS1 and PS2. A, qRT-PCR analysis of iPSCs and NPCs with the genotypes fPS1/fPS1;PS2 1/1 (WT), PS1 À/À ;PS2 1/1 (DPS1), and fPS1/fPS1;PS2 À/À (DPS2). Quantitative analysis shows the almost complete elimination of PS1 and PS2 in the DPS1-derived and DPS2-derived cells, respectively. Specific genes encoding components of the Notch signaling pathway (NOTCH1, NOTCH2, HES1, and HES5) were robustly increased in NPCs throughout neural patterning. The levels of both HES1 and HES5 were slightly but not significantly decreased in DPS1-NPCs compared with those in WT NPCs, whereas the levels of NOTCH1 and NOTCH2 were comparable among the genotypes. Data represent the mean 6 SEM (n = 3-5 of independent culture batches). ns, not significant; ***p , 0.001 by Tukey's test among the three genotypes in the NPC group. B, Experimental scheme of the colony-forming assay using fPS1/fPS1;PS2 1/1 and fPS1/fPS1;PS2 À/À iPSCs by flow cytometry. C, Quantitative analysis of the first neurospheres showing mCherry fluorescence. The fluorescence percentage was measured by flow cytometry after "partial" infection with lentivirus expressing mCherry-DCre or mCherry-Cre in fPS1/fPS1;PS2 1/1 and fPS1/fPS1; PS2 À/À neurospheres (NS; flow cytometry plots were shown in Extended Data Fig. 3-1). Data represent the mean 6 SD (n = 3 assays). ns, not significant by Tukey's test among the genotypes. D, Quantitative analysis of the second neurospheres showing mCherry fluorescence. The fluorescence percentage was measured by flow cytometry and then calculated as the % reduction in mCherry 1 cells from the first NS to the second NS (flow cytometry plots were shown in Extended Data Fig. 3-1). Data represent the mean 6 SD (n = 3 assays); ***p , 0.001 by Tukey's test among the genotypes.

Minimal developmental effect of PS inactivation on mature human neurons
To investigate whether the absence of PS1, PS2, or both affects neuronal differentiation in iPSC-derived cortical neurons, we first developed a cKO cortical neuronal culture system, in which PS1 can be eliminated by Cre transduction at any time. To circumvent any possible effects of PS deficiency during neural progenitor maintenance (Fig. 3), we infected fPS1/fPS1;PS2 1/1 and fPS1/ fPS1;PS2 À/À cells with lentivirus expressing Cre 5-6 d after terminal differentiation (day 20). When infected at a MOI = 3, ;90% of cells were positive for fluorescence in cortical neurons at day 45 (Fig. 4A). To determine exactly when PS1 was eliminated in this system, we performed Western blot analysis using fPS1/fPS1;PS2 1/1 neuronal cell lysates at 5-85 d after infection with the lentivirus expressing DCre or Cre. The levels of PS1 proteins were robustly decreased at 15 d after Cre transduction (day 35), and the extent of reduction progressively increased (Fig.  4B), whereas the levels of APP protein were relatively comparable. Intriguingly, the level of PS1 and PS2 proteins was gradually changed in DCre-infected cultures during neuronal maturation period (day 25-105) regardless of insufficient trials (Fig. 4B), consistent with prior report showing that PS is expressed differentially during mouse brain development (Kumar and Thakur, 2012). We quantified PS protein levels 25 d after Cre transduction (day 45) and found that the PS1 level was significantly reduced by ;80% compared with that in DCre-infected cultures (Fig. 4C). Interestingly, the PS2 level was significantly increased by ;25% in Cre-infected cultures, suggesting the presence of compensation for PS1 deficiency. To investigate whether the loss of PS affects neuronal differentiation in the cultures, we next performed immunocytochemistry using MAP2 antibody and measured neurite length of fPS1/fPS1;PS2 1/1 ;DCre (control), fPS1/fPS1; PS2 1/1 ;Cre (PS1-null), fPS1/fPS1;PS2 À/À ;DCre (PS2-null) and fPS1/fPS1;PS2 À/À ;Cre (PS1/PS2-null) neurons at day 45, in which lentiviruses were infected 5 d after terminal differentiation. Neurite length was almost comparable among the genotypes, despite a slight decrease in Cre-infected neurons (Fig. 4D), indicating negligible effect of PS absence on neuronal morphology in fully differentiated neurons. We further performed qRT-PCR of several neuronal markers. PS1 and PS2 mRNAs were almost absent in the neuronal cultures devoid of PS1 (PS1-null and PS1/ PS2-null) and PS2 (PS2-null and PS1/PS2-null), respectively, whereas the levels of all neuronal markers, including MAP2, synaptophysin, and PSD95, were comparable among the genotypes (Fig. 4D). Together, these results demonstrated that specific elimination of PS can be accomplished without any gross defects in neuronal differentiation.
Disturbed APP processing in PS-deficient human neurons Ab peptides are generated by sequential cleavage of the APP protein by b -secretase and g -secretase (De Strooper et al., 1998;Vassar et al., 1999). To investigate how the elimination of PS1, PS2 or both affects Ab production in iPSC-derived cortical neurons, we first performed sandwich ELISAs specific for Ab 40 and Ab 42 using culture medium from neuronal cultures at day 45. Surprisingly, the levels of Ab 40 were decreased only in PS1/PS2-null neuronal cultures, whereas cultures with other genotypes showed comparable levels of Ab 40 (Fig.  5A). Another independent iPSC line harboring fPS1/fPS1; PS2 1/1 also exhibited no decrease in Ab in the absence of PS1 alone (Extended Data Fig. 5-1A), excluding the possibility of clonal variability. Furthermore, the levels of Ab 42 were decreased in both PS2-null and PS1/PS2-null neuronal cultures but not in PS1-null cultures, suggesting the importance of PS2/g -secretase in Ab 42 production (Fig. 5A). Notably, the ratio of Ab 42/Ab 40 was not increased in PS1-null and/or PS2-null neuronal cultures (Extended Data Fig. 5-1B), unlike the case of FAD-linked PS1 mutations (Yagi et al., 2011;Woodruff et al., 2013;Kondo et al., 2017;Ishikawa et al., 2020;Sho et al., 2020). Inhibition of g -secretase with DAPT abolished the secretion of Ab 40 and Ab 42 (Extended Data Fig. 5-1C), indicating the successful measurement of g -secretasemediated Ab production; however, low levels of Ab were generated even with DAPT treatment, which is supposedly an APP cleavage product by a g -secretase-independent proteolysis (Ladror et al., 1994;Cataldo et al., 1995). These results demonstrated that inactivation of PS1 alone is not enough to eliminate Ab production in human cortical neurons, which is inconsistent with previous reports showing that Ab 40 and Ab 42 are significantly reduced in PS1 cKO mice (Yu et al., 2001;. We further examined another index of APP processing in PS-deficient human neurons. Ab peptides were produced directly from APP-CTFs, which are products cleaved by b -secretase and are the direct substrate of g -secretase. Therefore, the accumulation of APP-CTFs is a good indicator of g -secretase impairment. We performed Western blotting using an antibody for the APP Cterminal region and found that APP-CTFs were accumulated only in the lysates from PS1/PS2-null neuronal cultures (Fig. 5B), which is consistent with the results of the Ab ELISAs. We further performed immunocytochemistry of APP-CTFs, and again, the signals were robustly increased only in PS1/PS2-null neurons (Fig. 5C). These results prompted us to test whether PS1 elimination is effective enough in producing subsequent phenotypes in PS1-null neurons. When we used cortical neurons derived from PS1 KO (DPS1) iPSCs in which the PS1 gene was already knocked out, an accumulation of APP-CTFs was not observed (Extended Data Fig. 5-1D), demonstrating that a lack of PS1 alone is not enough to compromise APP processing. Next, we examined another substrate of g -secretase, N-cadherin, which is expressed in mature neurons as an essential hemophilic adhesion molecule at synapses (Fannon and Colman, 1996;Uchida et al., 1996). N-cadherin is subjected to sequential cleavage by ADAM10 and g -secretase (Marambaud et al., 2003;Reiss et al., 2005;Uemura et al., 2006). Ncad-CTF1, a cleaved product of N-cadherin produced by ADAM10, was significantly accumulated in both PS1-null and PS1/PS2-null neurons according to Western blot analysis (Fig. 5D). These results revealed that Cre-mediated PS1 elimination in this system is sufficient to assess the effects on substrate processing in human cortical neurons. Together, Because of posttranslational modification such as glycosylation, full-length of APP proteins exhibits two bands ;100 kDa in size. C, Western blot analysis of PS1-NTF, PS2-CTF, and b -actin. Representative blots are shown for each protein in fPS1/fPS1 iPSC neurons infected with DCre or Cre lentivirus. Data represent the mean 6 SEM (n = 3 of independent culture batches); **p , 0.01, ***p , 0.001 by Student's t test between the genotypes. D, Quantification of MAP21 neurite length at 45 d in fPS1/fPS1;PS2 1/1 and fPS1/fPS1;PS2 À/À iPSC neurons with infected with DCre or Cre lentivirus showed no alteration among the genotypes, despite a slight shorter tendency in PS1-deficient neurons. Data represent the mean 6 SD (n = 5 of independent culture batches). ns, not significant by Dunnett's test versus the control. E, qRT-PCR analysis at 45 d in fPS1/fPS1;PS2 1/1 and fPS1/fPS1;PS2 À/À iPSC neurons infected with DCre or Cre lentivirus showed the almost complete elimination of PS1 and PS2 in Cre-infected neurons and PS2 À/À neurons, respectively, whereas the levels of neuronal markers such as MAP2, synaptophysin, and PSD95 were comparable among the genotypic groups. Data represent the mean 6 SEM (n = 3-4 of independent culture batches). ns, not significant; ***p , 0.001 by Dunnett's test versus the control. Figure 5. Cleavage of APP and N-cadherin in the iPSC neurons of fPS1/fPS1;PS2 1/1 and fPS1/fPS1;PS2 À/À infected with DCre or Cre lentivirus. A, Specific ELISA measurement of Ab 40 and Ab 42 in the iPSC neurons of fPS1/fPS1;PS2 1/1 and fPS1/fPS1;PS2 À/À infected with DCre or Cre lentivirus. Quantitative analysis revealed a reduction in the Ab levels in the iPSC neurons of fPS1/fPS1; PS2 À/À infected with Cre lentivirus. Data represent the mean 6 SEM (n = 4-6 of independent culture batches); *p , 0.05, ***p , 0.001 by Dunnett's test versus the control. No change of Ab generation in the absence of PS1 alone was corroborated with the iPSC-neurons derived from another fPS1/fPS1;PS2 1/1 clone (#249-3; Extended Data Fig. 5-1A). DAPT treatment nearly abolished the generation of Ab 40 and Ab 42 (Extended Data Fig. 5-1C). B, Levels of APP-CTFs were quantified by Western blotting. Quantification analysis shows a massive increase in APP-CTFs in the iPSC neurons of fPS1/fPS1;PS2 À/À infected with Cre lentivirus. No accumulation of APP-CTFs in human neurons devoid of either PS1 or PS2 alone was further corroborated with the iPSC neurons from WT, DPS1, and DPS2 (Extended Data Fig. 5-1D). Data represent the mean 6 SEM (n = 4-5 of independent culture batches). ns, not significant; ***p , 0.001 by Dunnett's test versus the control. C, Levels of the APP-CTFs were quantified by immunocytochemistry. Quantification analysis showed a massive increase of the APP-CTFs in the iPSC-neurons of fPS1/fPS1;PS2 À/À infected with Cre lentivirus. Data represent the mean 6 SEM (n = 4-6 of independent culture batches). ns, not significant; **p , 0.01 by Dunnett's test versus the control. D, Levels of N-cadherin-CTF1s were quantified by Western blotting. In contrast to APP cleavage, N-cadherin was cleaved exclusively by PS1/g -secretase. Data represent the mean 6 SEM (n = 4-5 of independent culture batches). ns, not significant; *p , 0.05, ***p , 0.001 by Dunnett's test versus the control. E, Levels of phosphorylated (AT8) and total tau (Dako) were quantified by Western blotting. Most tau proteins in iPSC-derived neurons were phosphorylated (black our novel human neuronal system clarified the substrate specificity of PS1 and PS2. Considering that Ab has a causative relation with other pathologic lesions in the course of AD pathogenesis, we next examined whether acute ablation of PS1 and/or PS2 affects phosphorylation state of tau, another pathologic hallmark in AD patient's neurons. Unexpectedly, tau phosphorylation was not increased in PS-deficient neurons compared with PS intact neurons, indicating that tau phosphorylation is not directly regulated by PS/g -secretase (Fig. 5E).
As several amino acid residues are different between the human and mouse PS2 protein (Levy-Lahad et al., 1995;Rogaev et al., 1995;Vito et al., 1996), we first hypothesized the critical differences in PS2/g -secretase activity per se between humans and mice, which are attributed to inconsistent APP cleavage in PS1-deficient cells. To elucidate whether human and mouse PS2/ g -secretase exhibit specific APP processing in the same PS-deficient cellular context, we performed a PS2 complementation assay using PS-deficient mouse fibroblast cells, which stably express Swedish mutant APP (Fig.  6A,B). When we expressed human PS1 exogenously using an EF1a promoter-driven lentivirus along with an IRES-mediated Venus fluorescent protein (Nagai et al., 2002), Ab peptides were robustly generated in the culture medium, in which DAPT almost completely abolished Ab secretion (Fig. 6C). We next expressed human and mouse PS2 and measured Ab . PS2 from both species could secrete Ab peptides at equivalent levels when normalized to the expression levels of the Venus protein (Fig. 6C). Interestingly, we found that PS2 significantly increased the Ab 42/Ab 40 ratio compared with PS1, although there was no large difference in Ab secretion between PS1 and PS2. Using this exogenous expression system, we demonstrated that PS2/g -secretase itself does not exhibit species differences in its activity, and PS2-directed activity generated a longer form of Ab compared with PS1-directed activity; the latter findings are consistent with those of the Ab ELISAs in PS-deficient human neurons (Fig. 5A).

Unique subcellular localization of distinct c-secretase complexes containing PS1 or PS2 in human neurons
Previous reports showed differences in the subcellular localization of g -secretase complexes depending on their distinct subunit composition in human nonneuronal cell lines and mouse primary neurons, where the fluorescent protein-tagged g -secretase component was expressed heterologously (Meckler and Checler, 2016;Sannerud et al., 2016). To elucidate the subcellular localization of endogenous PS1 or PS2/g -secretase complexes in human neurons, we used a monoclonal antibody specific for glycosylated nicastrin (A5226A), which resides only in the active g -secretase complex (Hayashi et al., 2012). Only a small number of subunits are involved in the assembly of the g -secretase complex (Thinakaran et al., 1997;Kimberly et al., 2002); therefore, this antibody is useful to distinguish the nicastrin subunit in the g -secretase complex from free nicastrin. We first performed immunocytochemistry and found that both the areas and the number of puncta positive for the g -secretase complex were significantly reduced in PS1/PS2-null neurons compared with those in control neurons (Fig. 7A-C), indicating that active g -secretase complex does not form in the absence of PS. However, some signals were detected even in PS1/ PS2-null neurons, albeit with a robust reduction in the number of puncta, suggesting that a small proportion of nicastrin monomer can be recognized or that the g -secretase complex formed before Cre-lentivirus infection still remains in the PS1/PS2-null neurons. We next examined the colocalization of g -secretase and intracellular organelles using antibodies against EEA1 and LAMP1 as early endosome and late endosome/lysosome markers, respectively. The percentages of g -secretase complexes costained with EEA1 and LAMP1 were 4.3 6 0.6% and 40.4 6 8.5%, respectively (Fig. 7;, demonstrating that approximately half of the g -secretase complex exists in LAMP11 organelles in human neurons. Interestingly, colocalization signals for g -secretase and LAMP1 were significantly decreased by ;60% and ;80% in PS2-null and PS1/PS2-null neurons compared with those in controls, respectively (Fig.  7D), although the area of LAMP11 organelles was comparable among the four genotypes (Fig. 7E). These results are consistent with a recent report that PS2 is mainly localized in late endosomes/lysosomes in nonneuronal cells and rodent neurons (Meckler and Checler, 2016;Sannerud et al., 2016). Together, these results strongly suggest that PS2/ g -secretase localizes largely in LAMP1 organelles and that almost no g -secretase stays in EEA11 organelles in human neurons.
Endosome enlargement is detected in AD patient brains and AD model-derived neurons as a characteristic cytopathology (Cataldo et al., 2000;Kwart et al., 2019). Because the impairment of proper APP processing and the concomitant accumulation of APP-CTFs are culprits for this phenomenon, we analyzed EEA11 puncta in human neurons. Surprisingly, no enlargement in EEA11 signals was observed in PS1/PS2-null neurons (Extended Data Fig. 7-1), despite the robust accumulation of APP-CTFs in these neurons (Fig. 5B,C). Therefore, it is most likely that a loss of PS causes an accumulation of APP-CTFs ( Fig. 5B) but not an enlargement of early endosomes, which probably necessitates additional impairment, such as an APP/PS1 mutation, in addition to the accumulation of APP-CTFs (Kwart et al., 2019).

Discussion
The use of gene KO methodology in animals or cultured cells is a conventional means to assess physiological continued arrowhead) rather than unphosphorylated form (gray arrowhead) using total tau antibody, similar to embryonic brains. The state of tau phosphorylation was not altered by the absence of PS1 and/or PS2. Data represent the mean 6 SEM (n = 3-4 of independent culture batches). ns, not significant by Dunnett's test versus the control. and/or pathophysiological functions of the gene of interest. However, a simple KO strategy has sometimes been hampered by developmental disturbance: one such example is the PS1 gene, the germline KO of which causes perinatal death in mice Wong et al., 1997). To circumvent this developmental lethality in mice, both PS1 cKO and PS cDKO mice have been created by crossing floxed PS1 mice with aCaMKII-Cre transgenic mice, along with or without PS2 KO mice, demonstrating that PS is essential for cortical neuron survival and synaptic functions in mice (Saura et al., 2004;Zhang et al., 2009). In this study, to investigate the physiological functions of PS in the human neural cell context, we developed a novel human iPSC-derived model in which PS1 is ablated with the Cre/loxP system to avoid any possible developmental impediment. Indeed, PS1 cKO iPSCs can circumvent the maintenance deficit of PS1/PS2-null NPCs. By examining mature human neurons lacking PS, we clearly revealed the human-specific regulation of PS/ g -secretase and substrates (APP and N-cadherin) and  . Subcellular localization of PS1/g -secretase complexes and PS2/g -secretase complexes. A, Representative images of iPSC neurons stained with antibodies specific for g -secretase complexes (A5226A), LAMP1 and MAP2 are shown. There are many puncta of g -secretase complexes and LAMP1 in the perinuclear region and neurites. Arrows indicate colocalization puncta. Scale bar: 2 mm. The similar experiments using EEA1 antibody were shown in Extended Data Figure 7-1. B, C, Quantification of the g -secretase complex puncta area (B) or number (C) per lentivirus-infected neuron. Data represent the mean 6 SEM (n = 5-7 of independent culture batches). ns, not significant; *p , 0.05, **p , 0.01 by Dunnett's test versus the control. D, Colocalization of g -secretase complexes and LAMP1. Quantification of these signals shows a decrease in LAMP1 colocalization in the absence of PS2. Data represent the mean 6 SEM (n = 5-7 of independent culture batches). ns, not significant; *p , 0.05, **p , 0.01 by Dunnett's test versus the control. E, Quantification of the LAMP1 puncta area in lentivirus-infected neurons. Data represent the mean 6 SEM (n = 5-7 of independent culture batches). ns, not significant by Dunnett's test versus the control. compared them with previous murine-based results (Yu et al., 2001;Watanabe et al., 2014). The discrepancy may result from the inherent intracellular circumstances but not from primary sequence differences in the PS protein.
Using a specific antibody against mature glycosylated nicastrin, we characterized the subcellular localization of nearly half of the endogenous PS/g -secretase complex in late endosomes/lysosomes, where Ab 42 is relatively abundantly generated. These findings are the first to show the physiological function/location of an endogenous but not heterologously expressed PS/g -secretase complex in a human neural cellular context. g -Secretase is composed of four integral subunits, PS, nicastrin, Aph-1, Pen-2 (Kimberly et al., 2003;Takasugi et al., 2003), and only one molecule from each subunit is assembled into the complex (Sato et al., 2007). Importantly, PS and Aph1 each have a homolog; PS1, PS2, Aph-1A (which further forms L and S variants by alternative splicing), and Aph-1B can form six putative g -secretase complexes. Furthermore, g -secretase cleaves many substrates, almost all of which are Type I transmembrane proteins such as APP and Notch, leading to nuclear signaling that results in transcription and proteostasis of membrane proteins (Haapasalo and Kovacs, 2011). One important unsettled question is whether distinct g -secretase complexes can equally cleave their different target substrates. Previous reports have also revealed that there is relative substrate specificity among distinct g -secretases in nonneuronal cell cultures and in mice (Serneels et al., 2009;Sannerud et al., 2016). In the present study using human neural cells, we demonstrated the exclusive specificity of endogenous PS1/g -secretase toward Ncadherin, although such clear specificity has not been concluded in mouse fibroblasts (Marambaud et al., 2003). In this regard, it is most likely that PS1/g -secretase performs Ncadherin cleavage strictly at the site lacking a PS2/g -secretase in human neurons. However, it remains to be determined whether diverse g -secretase complexes containing Aph-1A or Aph-1B show any substrate specificity in human neurons.
g -Secretase and its substrates are both membrane proteins; thus, proteolytic reactions occur only in the same subcellular compartment. Indeed, several groups have demonstrated that the different subcellular localizations of each g -secretase underlie their substrate specificity in nonneuronal cells (Tarassishin et al., 2004;Meckler and Checler, 2016;Sannerud et al., 2016). Sannerud and colleagues showed that PS2/g -secretase is localized in late endosomes/lysosomes via the specific targeting signal of PS2 (Sannerud et al., 2016), which is consistent with our results in human neurons. However, in this study, a small proportion of g -secretase still resided in LAMP11 organelles in the absence of PS2. This may result from the compensatory expansion of PS1 in their location because of a loss of PS2. Alternatively, LAMP1 immunoreactivity is found more broadly in locations beyond the late endosome/lysosome, according to the recent literature (Cheng et al., 2018). In contrast, we could barely detect colocalization between EEA11 organelles and g -secretase, suggesting that EEA11 early endosomes are a transient location of g -secretase (Kanatsu et al., 2014). Furthermore, in relation to its exclusive N-cadherin cleavage (Fig. 5D), PS1/g -secretase but not PS2/g -secretase could be targeted in the plasma membrane, where ;6% of total g -secretase exists, as shown in cell lines (Chyung et al., 2005). Other membrane organelles, including the Golgi apparatus and recycling endosome, remain to be analyzed in the future.
In terms of APP processing, we and other groups have shown obvious species-specific or cell type-specific differences. Using PS KO cells, APP can be efficiently processed by both PS1 and PS2 in some systems (Lessard et al., 2019;Pimenova and Goate, 2020), including those examined in this study, whereas PS2 enzymatic activity is not efficient in others Arber et al., 2019). This discrepancy might be a result of distinct PS2 expression levels in each system as well as differences in subcellular localization, as discussed above. Conceivably, this discrepancy could be partly caused by differences between neuronal and other neural cells. The elimination of PS1 in excitatory neurons of PS1 cKO mice could affect other cell types, leading to non-cell autonomous effects on APP processing in neurons. Surprisingly, Woodruff et al., showed that human isogenic iPSCs carrying the PS1 DE9 mutation, which is a partial loss-of-function mutation, led to an accumulation of APP-CTFs despite the presence of a total of three WT PS1/PS2 alleles (Woodruff et al., 2013). The difference in APP processing in iPSC-derived human neurons might result from the usage of different iPSC lines or neuronal differentiation protocols between the other study and ours. Alternatively, DE9-mutant PS1/g -secretase could confer a putative dominant-negative effect on g -secretase bearing WT PS1/PS2 (Watanabe and Shen, 2017).
Likewise, the scenario is similar for Notch processing. PS1 is crucial for neural development during embryogenesis through Notch signaling, and PS1 KO homozygous mice exhibit a perinatal lethal phenotype Wong et al., 1997). In this study, however, the extent of the PS1 contribution to the maintenance of the human neurosphere was much less than that of the mouse neurosphere (Hitoshi et al., 2002). This discrepancy might result from sufficient compensation by PS2 in human PS1-deficient NPCs. Alternatively, as Notch pathway is modulated by other signaling pathways such as Sonic hedgehog, Bone morphogenetic proteins and Wnt (Gajera et al., 2010;Lin and Hankenson, 2011;Fujimori et al., 2017, we cannot exclude the possibility that other pathways could efficiently compensate the decreased Notch pathway in PS1-deficient NPCs. More interestingly, Arber and colleagues recently demonstrated that FAD-linked PS1 mutations bring about precocious characters in iPSC-derived NPCs (Arber et al., 2021). Despite some discrepancy between their study (PS1 mutation) and ours (PS deficiency), it is more likely that an impaired PS/g -secretase resulted in disturbance of NPC maintenance. Collectively, these results clearly demonstrate the contextual differences between g -secretase and its substrate, underscoring the importance of human neuronal models in the scrutiny of AD pathogenesis.
In this study, we clarified the heterogeneity of PS/ g -secretase complex in human cortical neurons, which underlies different production modes of endogenous Ab species by individual PS/g -secretase complex. Although our study would provide a unique cellular model to scrutinize the physiological production mechanism of specific Ab species in the context of human neurons, this PS cKO iPSC model did not always mimic the authentic model of AD pathogenesis. Expectedly, the complete loss of PS function in human cortical neurons showed the negligible generation of both Ab 40 and Ab 42 (Fig. 5A), which results in an erratic decrease of Ab 42/Ab 40 ratio (Extended Data Fig. 5-1B). Given that most FAD-linked PS mutations cause a partial loss of g -secretase function (Szaruga et al., 2015), especially carboxyl-peptidase activity, leading to an accumulation of toxic longer Ab 42 and Ab 43, the effect of FAD-linked PS mutations is more complicated in the course of AD pathogenesis, not the case of simple absence of g -secretase activity. As PS mutations show an autosomal dominant inheritance trait in FAD pedigree, it is plausible that mutant PS plays a dominant-negative role in WT PS/g -secretase complexes as shown previously (Heilig et al., 2013;Zhou et al., 2017). Future studies are needed to demonstrate that a physical proximity between WT and mutant PS/g -secretase is detected in the same organelle from this point of view. Moreover, acute PS elimination did not increase tau phosphorylation in human cortical neurons (Fig. 5E), another AD pathologic hallmark, while such anomalies were observed in iPSC-derived neurons from AD patients (Israel et al., 2012;Ochalek et al., 2017). This suggests that tau phosphorylation requires accumulation of toxic Ab species (Jin et al., 2011), leading to ultimate lesions such as synapse deficits and neuronal degeneration (Nieweg et al., 2015;Kouroupi et al., 2017;Gajera et al., 2019).
In summary, this study uncovered the distinct regulation of the PS/g -secretase complex in iPSC-derived human neural cells compared with that revealed in previously reported mouse studies, which may underlie the failures of clinical trials for g -secretase inhibitors. Despite the profound phenotypic differences of the PS/g -secretase used in this study, however, the detailed molecular mechanisms underlying the substrate cleavage and subcellular localization of distinct PS/g -secretases remain to be resolved. Furthermore, it would be much better to recapitulate our results in more suitable systems such as human brain organoids/spheroids, in which glial cell types can be developed concomitantly and interacted with neurons functionally (Paşca et al., 2015;Madhavan et al., 2018;Ormel et al., 2018). Using the novel cellular model in this study, future studies will clarify the causative molecular changes in PS/g -secretase from physiological to pathophysiological states in the course of AD pathogenesis and lead to the development of novel therapeutic medicines.