Early observational studies showed a compelling beneficial effect of estrogen on AD in cognitive tests and positron emission tomography scans using bolus injected [¹⁵O] water. Honjo et al. (1989) treated female patients with AD with oral conjugated estrogen (estrone sulfate as the main component) and suggested an improvement in cognitive function after 6-weeks of treatment (Honjo et al., 1989). Henderson et al. (1996) compared female patients with AD who received estrogen replacement therapy with female and male patients with AD who did not receive estrogen replacement therapy. They showed that female patients with AD who received estrogen showed significantly better performance on cognitive function tests than female and male patients who did not receive estrogen (Henderson et al., 1996). In addition, Resnick et al. (1997) showed that female patients who received HRT had better cognitive function tests, especially those related to visual memory (Resnick et al., 1997). A subsequent study also showed that HRT users had better regional cerebral blood flow (rCBF) in the right parahippocampal gyrus, right precuneus, and right frontal regions during memory tasks, suggesting better activation of the brain areas that are severely affected in patients with AD (Resnick et al., 1998).

Based on the positive results of the observational studies, RCTs were launched. However, the RCTs showed various results, including positive, partially positive, and negative effects of HRT (Table 1). Some of the first RCTs indicated that HRT had no effect on AD or adversely increased the risk of developing AD, contrary to expectations. Henderson et al. (2000) showed that the treatment of conjugated equine estrogens (CEE) for 4 weeks in postmenopausal female patients with mild to moderate AD did not result in significant improvement in cognitive function compared to the placebo group (Henderson et al., 2000). Mulnard et al. (2000) treated menopausal female patients with AD with high and low doses of CEE for 1 year, a longer period than previous studies, and found that CEE did not affect attenuating AD (Mulnard et al., 2000). One large clinical cohort, the Women’s Health Initiative Memory Study that included more than 4,000 menopausal female patients who were receiving CEE and/or medroxyprogesterone acetate for more than 4 years, also revealed negative or even adverse effects of hormone treatment. Treatment with estrogen and progesterone has been reported to have adverse effects that may promote the development of AD. Thus, the opposed estrogen therapy, which is a form of estrogen therapy supplemented with progesterone to prevent the adverse effects of excessive estrogen alone, such as enhancing the risk of endometrial cancer development, was less beneficial in preventing the development of AD (Shumaker et al., 2003; Craig et al., 2005).

Due to the negative results of RCTs during the early 2000s, the following RCTs focused more on the stratification of patient characteristics (e.g., age, cognitive status, APOE genotype, duration after menopause), along with the formulation of drugs and the route of administration (e.g., oral CEE, transdermal 17β-estradiol). These RCTs were finally able to assert the beneficial effects of estrogen in human clinical trials. Valen-Sendstad et al. (2010) showed that females without the APOE ε4 allele (a major genetic risk for AD) received estradiol and norethisterone, they had better mood and depression than females with the APOE ε4 allele (Valen-Sendstad et al., 2010). Silverman et al. (2011) compared the continued and discontinued HRT effects among cognitively normal postmenopausal females with a family history of AD and showed the following results: (i) 17β-estradiol had a better effect than CEE; (ii) unopposed estrogen therapy preserved more cortical metabolism compared to the opposed one; and (iii) females with longer endogenous estrogen exposure showed relatively preserved metabolism of specific brain areas more than females with shorter endogenous estrogen exposure (Silverman et al., 2011). Moreover, the Kronos Early Estrogen Prevention study, which included only recent postmenopausal females, concluded that 17β-estradiol could lower Aβ deposition, especially in those with APOE ε4 among the recent menopausal female population (Kantarci et al., 2016). In addition to conventional transdermal 17β-estradiol or oral CEE, selective estrogen receptor modulators as another form of medication are also being investigated for their effects on improving cognitive function. A study by Wang et al. (2020) showed that PhytoSERM [an estrogen receptor beta (ERβ) modulator comprised of genistein, daidzein, and S-equol] improved cognitive function, especially for verbal learning and executive function (Wang et al., 2020).

Similar to estrogen and progesterone, testosterone levels also decrease with age, and low testosterone levels contribute to cognitive decline in elderly males (Moffat, 2005; Lv et al., 2016). Although only a few RCTs have evaluated the effect of testosterone on cognitive function, studies of short-term testosterone replacement therapy in elderly patients with MCI or AD found improvement in cognitive function after treatment, especially for spatial and verbal memory (Tan and Pu, 2003; Cherrier et al., 2005), while long-term therapy of testosterone showed only modest or no significant improvement (Asih et al., 2015; Huang et al., 2016; Wahjoepramono et al., 2016). Meanwhile, Huang et al. (2015) reported no significant cognitive improvement after testosterone administration, even among females with low testosterone levels (Huang et al., 2015).

After the establishment of iPSCs technology by the group of Shinya Yamanaka (Takahashi and Yamanaka, 2006; Takahashi et al., 2007), human cell models of many diseases, including neurological disorders, have been generated using iPSCs (Okano and Yamanaka, 2014; reviewed in Imaizumi and Okano, 2014; Rowe and Daley, 2019; Sharma et al., 2020). For AD, iPSCs were first established from fibroblasts obtained from patients with familial Alzheimer’s disease (fAD) with mutations in PS1 (A246E) and PS2 (N141l) (Yagi et al., 2011), and the iPSCs were differentiated into neurons. These neurons induced by iPSCs recapitulated the phenotypes of AD, such as increased Aβ₄₂ levels and responded to AD targeting drugs, such as γ-secretase inhibitors and modulators. The following models of fAD with different mutations also showed similar phenotypes (reviewed in Zhang et al., 2016). In addition to the models of familial AD cases, a sporadic AD model of iPSC-derived neurons was established in 2012, in which increased Aβ levels, pTau protein, and other AD phenotypes were observed (Israel et al., 2012). Recent studies using iPSCs focused on sAD risk genes that increase the risk of developing AD, such as SORL1 (Young et al., 2015) and APOE. However, since the comparison of iPSCs from mutation carrier and non-carrier cannot exclude the effects of other SNPs that are different between the mutation carrier and non-carrier, using genome-editing technology like CRISPR/Cas9 allows us to introduce only the mutation of interest and examine the effects. Knupp et al. (2020) generated the SORL1 knockout iPSCs using CRISPR/Cas9 and could conclude that the loss of SORL1 induced early endosome enlargement (Knupp et al., 2020). Furthermore, another study utilized CRISPR/Cas9 to establish APOE ε3/ε3 and APOE ε4/ε4 hiPSC lines that have equal genomic background except for the APOE gene. Then, they investigated functions of APOE genotypes on diverse cell types including neurons, astrocytes, and microglia-like cells (Lin et al., 2018). Since the AD models established from iPSCs can reflect the phenotypes of the disease and respond to drugs, this tool can be used to investigate both cell-autonomous and non-cell-autonomous effects arising from sex differences within the AD population.

In our analyses of the neurons differentiated from human iPSCs of healthy female donor (1210B2 line) and female donor of fAD (APP V717L line) (Figure 1A) (Kondo et al., 2013; Nakagawa et al., 2014), secreted Aβ peptides were measured by enzyme-linked immunosorbent assay. Neurons from fAD donor increased Aβ_42/40, indicating that iPSC-derived neurons recapitulated the disease pathophysiology of AD (Figure 1B) (Jack et al., 2013; Lee et al., 2020). Furthermore, iPSC-derived neurons of the fAD donor increased the frequency of Ca imaging using Fluo-8 compared to healthy control (Figure 1C). This finding was reminiscent of the finding of Zott et al. (2019) that neuronal hyperactivation in AD mouse models is caused by Aβ-induced accumulation of perisynaptic glutamate (Zott et al., 2019). These neurons derived from iPSCs also responded to short-time exposure to 17β-estradiol. Exposure to 17β-estradiol for 15 min increased the Ca oscillation (Figure 1D). Previously, Zhang et al. (2010) also performed Ca imaging on human and mouse embryonic stem cell (ESC)-derived neurons after treatment with 17β-estradiol and showed increased neuronal firing (Zhang et al., 2010). Increased activation of neurons after acute treatment with estradiol is suggested to be mediated by the stimulation of L-type Ca²⁺ channels, which activate the MAPK/ERK pathway and promote the firing of neurons as part of the rapid signaling cascade (Sheldahl et al., 2008; Vega-Vela et al., 2017; Albert-Gascó et al., 2020). Shum et al. (2015) also treated iPSC-derived neurons from healthy donors with 17β-estradiol for 24 h and showed increased dendritic branching (Shum et al., 2015), suggesting promotion of neuronal microstructure by estrogen in the iPSC-derived models. The increased neuronal branching may explain the preserved cognitive function by HRT, as shown in previous RCTs. Further experiments such as the treatment of 17β-estradiol before/after Aβ treatment in iPSC-derived neurons will tell us whether 17β-estradiol protects neurons against oxidative stressor. Another study differentiated neurons from the iPSCs of the autism individual revealed an increased expression of androgen receptor and brain-derived neurotrophic factor after testosterone treatment, suggesting that the iPSC-derived neurons also respond to testosterone treatment (Adhya et al., 2018). These evidence suggest that the in vitro model using iPSCs responds to sex steroid hormone and has potential for future study of sex differences, including the effects of hormone therapy, among AD patients.

Induced pluripotent stem cells (iPSCs) can hold the genetic information of the donors, thus allowing in vitro study of diseases with the donor’s genetic background (Stadtfeld and Hochedlinger, 2010). Non-cell-autonomous factors, such as the exposure time to hormones and the amount of Aβ aggregates, can also be controlled. Moreover, the iPSC-derived disease model can recapitulate the phenotypes of individual patients and can be used for disease stratification and drug screening of incurable neurodegenerative diseases. Previously, stratification and drug screening for sporadic disease were performed in amyotrophic lateral sclerosis (ALS). Motor neurons differentiated from patients with ALS-derived iPSCs showed multiple phenotypes, such as neurite retraction, lactate dehydrogenase leakage, increased cleaved caspase-3, and abnormal protein aggregation. These phenotypes in vitro matched the clinical manifestations and could stratify the disease (Fujimori et al., 2018). The study further identified ropinirole, a well-known medication for PD, as a drug candidate that could be used to treat patients with ALS. This in vitro study finally led to a successful clinical trial of ropinirole in patients with ALS (Morimoto et al., 2019; Okano et al., 2020; Keio University School of Medicine, 2021). The success of ALS indicated the possibility of stratification and prediction of drug response for patients with AD using iPSC models derived from patients along with their clinical information. Additionally, one recent study also generated the human iPSCs-derived model of microglia and confirmed that female sex and APOE genotype drive the microglial transcription profiles, thus indicating another important aspect of sex difference and APOE genotype in AD (Moser et al., 2021).

The cell-autonomous effect of different sexes on cell models can be observed through several readouts under pathogen-free conditions, while non-cell-autonomous phenotypic changes can be observed by exposing the cells to pathogens or drugs. Furthermore, genetic information (especially AD risk genes) and clinical information of the patients can be included in the stratification of the disease. The effects of hormone therapy in patients can be predicted by phenotypic changes based on the results of the iPSC model. Finally, stratification for hormone treatment using iPSCs from individual participants would yield a robust simulation before the clinical trial, although the cost and time for the generation of iPSCs and subsequent analysis of cell models from individual participants will be a barrier (Figure 1E). Currently, the development of the 3D model such as brain organoids has advanced the disease modeling using iPSCs and were able to recapitulate the patient’s pathophysiology (Chiaradia and Lancaster, 2020). For example, the 3D brain organoid could recapitulate the amyloid aggregation which 2D model could not (Gonzalez et al., 2018). In 2D culture, the secreted Aβ is released into the medium and cannot be concentrated high enough to aggregate. However, in 3D culture, the secreted Aβ can be constrained in the extracellular space of the brain organoids and can increase the concentration high enough to aggregate. Furthermore, the utilization of 3D organoid in combination with the co-culture with other CNS cells/structure like microglia and blood vessels would yield a better modeling of disease and can be used for observation of the hormone therapy in the future.

Nevertheless, the iPSC technology has some limitations such as the genetic alterations that could occur during the reprogramming of the somatic cells to the iPSCs. Even though new reprogramming methods using the non-integrating tools like Sendai virus and episomal vectors caused less damages than the first generation of the reprogramming with lentivirus (Kang et al., 2015), the genetic alterations can still occur. Thus, many of the tests such as karyotyping, genotyping, stem cell markers’ expressions are usually required to confirm the quality of the established iPSCs before using for the modeling of diseases. In addition to neurons differentiated from iPSCs, human neurons can also be induced directly from somatic cells, such as fibroblasts (Victor et al., 2018). Directly induced neurons (iNs) obtained from the fibroblasts preserved the aging status of the donors, while the aging status of the donor was erased in iPSCs. The iNs of sporadic AD patients could also reflect AD phenotypes, such as uncomplicated dendrites, reduced synapses, epigenetic erosion, and increased DNA damage (Mertens et al., 2021). Thus, AD models of iN cells can also be used to evaluate aging-dependent sex effects and could provide insights into sex and age-related changes in AD.