All experiments followed the University of Illinois at Chicago Animal Care Committee protocols. EFAD mice express five familial AD (FAD) mutations and human APOE. Two groups of EFAD (5×FAD^+/-/human APOE ^+/+) mice were used: female E3FAD and female E4FAD mice (24). The mice were ear-tagged during genotyping, and the investigators were blinded about APOE, treatment, and age. We initially designed this study to test the interactive effect of APOE and treatments (sham, OVX -/+ E₂) on learning/memory and Aβ pathology in EFAD mice. However, due to the COVID-19 pandemic, we had to restructure the mouse enrollment for this study due to personnel restrictions and perform the surgeries and behavioral experiments for the E3FAD and E4FAD mice separately.

EFAD mice were acclimatized to plain hydrogel in place of drinking water 3 days prior to OVX. Bilateral OVX or sham surgery was performed on female E3FAD and E4FAD mice as described previously (25, 26) at either 4 or 8 months of age. Immediately after OVX, the mice were treated with hydrogel with or without 13.9 μg/mL β-estradiol (E₂). Each mouse consumes ~4.5 mL hydrogel/day, resulting in a dose of 2.5 mg/kg/day that was selected based on previous studies (27–29). The mice were treated from 4 to 8 months or from 8 to 12 months of age.

In the week prior to sacrifice, mouse behavior was tested using a modified Morris water maze (MWM) protocol (30). A four-trial shaping procedure was conducted 1 day prior to testing, during which the mouse was placed in various locations within the pool area (i.e., on a platform, near the platform, between the platform and a ring, and near the edge of the ring) to habituate the mice. After the shaping trials, mouse behavior was tested in acquisition trials for five consecutive days consisting of 4 × 1 min trials/day with latency to the platform recorded for each trial. After the last acquisition on day 5, a single probe trial was run with the platform removed, and the readouts included latency to platform and latency to target quadrant (previously described (31–33)). Both acquisition and probe trials were recorded and analyzed with ANY-maze software (Stoelting Co., Wood Dale, IL, USA).

At the end of the study, vaginal smear was used to determine whether the mouse was in proestrus/estrus or metestrus/diestrus (34, 35). The mice were then anesthetized with ketamine (100 mg/kg) and xylazine (5 mg/kg) via intraperitoneal injection and perfused. Uterine horns were dissected from the mice and weighed. Then, the brains were removed and harvested for biochemical and immunohistochemical analysis as described previously (36, 37). For biochemical analysis, the cortex was dissected from the hemi-brain, flash-frozen in liquid nitrogen, and then stored at -80°C. The hemi for immunohistochemical analysis was drop-fixed in 4% paraformaldehyde for 24 h and then transferred to phosphate-buffered saline containing 0.01% sodium azide until ready to be sectioned on a sliding microtome.

Frozen cortices dissected from the mouse hemi-brains were homogenized in 70% formic acid at 1 mL/150 mg brain tissue and mixed by end-over-end rotation for 2 h at room temperature with vortexing. The samples were then centrifuged (100,000 × g, 1 hour at 4°C), and the formic acid-soluble fraction was neutralized (with 20 volumes of 1 M Tris base), aliquoted, and frozen at −80°C (38). Total protein in the formic-acid-soluble extracts was quantified using the Bradford assay, and formic-acid-soluble Aβ42 was measured by ELISA following the manufacturer’s instructions (24, 38, 39). A list of all the antibodies used is provided in Supplementary Table S1 .

Serial sagittal brain sections (35 μm thick, 280 μm apart, ~0.24–3.44 mm lateral) from EFAD mice were immunostained for Aβ deposition (24, 33, 40). The stained sections were imaged at ×10 magnification with a Zeiss fluorescence microscope and analyzed for cortical area covered by MOAB-2 in the cortex using ImageJ by investigators blinded to treatment. A list of all the antibodies used is provided in Supplementary Table S1 .

Supplementary File S1 (Data Sheet 1) is a Word file containing one table and two figures. Supplementary File S2 (Data Sheet 2) is an Excel file containing all raw data including the number of mice and statistical analysis. MWM acquisition data was analyzed by repeated-measure univariate general linear model for within-subject effects (independent variable: day and treatment). All other statistical analyses were conducted using univariate general linear models for between-subjects effects with treatment as the independent variable. All statistical tests were followed with Bonferroni’s post-hoc tests in in SPSS (IBM SPSS Statistics for Macintosh, Version 29.0.1.1); p < 0.05 was considered significant. Data are presented as scatter plots with the mean and standard error of the mean (SEM).

APOE3 is often used as a control group in APOE research in comparison with APOE4. Thus, we initially focused on the impact of OVX and E₂ treatment on behavior and pathology in E3FAD mice due to the important implications for a large proportion of AD patients (19).

OVX results in disruption of uterine horn weight and estrous stage/cycling in vivo (45–48). Therefore, we first confirmed that OVX induced uterine atrophy and the effect of E₂. As expected, in E3FAD mice, OVX decreased uterine horn weights by 43% compared to sham (p = 0.07), and E₂ increased uterine horn weights by ~100% and 250% compared to sham mice and the OVX group, respectively ( Figure 1A ). We next evaluated the impact of OVX and E₂ on estrous stage distribution among the mice. In general, mice in proestrus and estrus stages are associated with higher levels of circulating E₂ in the periphery compared to mice in metestrus and diestrus stages. We confirmed that OVX decreased the proportion of mice in proestrus/estrus compared to sham group, and E₂ treatment after OVX increased the proportion of mice in proestrus/estrus compared to OVX and sham group ( Figure 1B ; Supplementary Figure S2 ).

Cognitive decline is one of main symptoms of AD (49). In FAD transgenic mouse models, including EFAD mice, learning and memory deficits are often assessed using the Morris water maze test (31, 32, 50, 51). In E3FAD mice, during acquisition phase, there was a main effect of training day but not treatments ( Figure 1C ). However, it appeared visually that E₂-treated mice had better performance on day 5 compared to other groups. Indeed on day 5 the latency to platform was significantly lower in E₂ group compared to the OVX group. Thus, although learning was not affected by OVX, E₂ treatment marginally improved learning in E3FAD mice. Next, we evaluated memory in a single probe trial. We found that probe measures were impacted by both OVX and E₂. Indeed both latency to platform and target quadrant followed the order: OVX > Sham ~ E₂. Thus, in E3FAD mice, OVX resulted in memory deficits, which were mitigated by E₂ ( Figure 1D ).

Extracellular Aβ is a major pathological hallmark and diagnostic criteria of AD in humans (52) and may be modulated by OVX and E₂ treatment as found in FAD mice (8, 53). Therefore, we next evaluated the impact of OVX and E₂ on Aβ levels in E3FAD mice using biochemistry and immunohistochemistry ( Supplementary Figure S2 ). Surprisingly, we found that both OVX and E₂ treatment mice had significantly decreased levels of formic-acid-soluble Aβ compared to sham mice in E3FAD ( Figure 1C ). Consistent with those results, OVX significantly decreased Aβ deposition compared to sham mice ( Figure 1E ), and E₂ lowered cortical Aβ (p = 0.07) compared to sham ( Figure 1F ).

Collectively, these data demonstrate that, compared to sham surgery, OVX resulted in uterine horn atrophy, changes in estrous stage distribution, memory deficits, and lower Aβ levels in E3FAD mice. E₂ mitigated the detrimental effect of OVX on the uterine horn, estrous stage distribution, and memory, with no effect on Aβ levels.

AD risk is higher in APOE4 carriers compared to APOE3 carriers, particularly female individuals (19, 20). However, there is limited in vivo data on how APOE4 modulates the effect of OVX and E₂ on behavior and Aβ pathology. Therefore, we next investigated the impact of OVX and subsequent E₂ treatment in E4FAD mice. OVX decreased uterine horn weight by ~32% in E4FAD mice compared to the sham group ( Figure 2A ). Furthermore, E₂ treatment resulted in uterine hypertrophy with ~100% to 250% increase in uterine horn weight compared to mice that underwent either sham or OVX surgery, respectively ( Figure 2A ). We also confirmed that OVX decreased the proportion of mice in proestrus/estrus, compared to the sham group, and that E₂ treatment increased the proportion of mice in proestrus/estrus after OVX ( Figure 2B ; Supplementary Figure S1 ). In terms of behavior, we found that, during acquisition phase, there were no main effects of either training day or treatments ( Figure 2C ). There was also no effect of OVX or E₂ treatment in probe trial measures ( Figure 2D ). Although OVX and E₂ did not impact insoluble Aβ42, OVX increased Aβ deposition (compared to sham mice), which was lowered by E₂ ( Figures 2E, F ; Supplementary Figure S2 ).

Taken together, OVX impacted uterine horn weights and estrous stage distribution and increased cortical Aβ deposition without affecting MWM readouts in E4FAD mice. E₂ attenuated the effect of OVX on uterine horn weights and estrous stage distribution and decreased cortical Aβ deposition, with no effect on behavior.

Our data demonstrated in E3FAD mice that OVX at 4 months of age is detrimental, and E₂ treatment from 4 to 8 months of age may be protective for behavior. As described above, that paradigm was selected based on pathology. We next asked whether E₂ would be beneficial if OVX was performed at an older age with greater Aβ pathology. Therefore, we focused on the effects of E₂ treatment from 8–12 months of age (OVX at 8 months) on behavior and Aβ pathology in E3FAD mice.

In E3FAD mice, compared to sham, OVX at 8 months of age decreased uterine horn weights by 32% ( Figure 3A ) and increased the proportion of mice metestrus/diestrus. Furthermore, E₂ increased uterine horn weights by ~200% and 400% compared to OVX and sham mice, respectively, and increased the proportion of mice in proestrus/estrus ( Figure 3B ; Supplementary Figure S1 ). However, we found 1/10 OVX mice in estrus stage. Although it seems impossible, neonatal treatment of female mice with estrogen or androgen has been demonstrated to induce ovary-independent persistent proliferation and cornification of vaginal epithelium that may result in the classification of the mice as under estrous phase (54). Therefore, it may be a one-off cytological presentation. Thus, E₂ mitigated late OVX-induced changes in both uterine horn weight and estrous stage distribution in E3FAD mice. In terms of behavior in MWM, during acquisition trials, there was a main effect of treatments ( Figure 3C ). The post-hoc analysis revealed that latency to platform in acquisition trials was demonstrated by the sham ~ OVX > E₂ group. However, both OVX and E₂ did not affect the probe trial measures ( Figure 3D ). We also found that, after late OVX, neither OVX nor E₂ impacted cortical insoluble Aβ42 ( Figure 3E ) or Aβ deposition ( Figure 3F ; Supplementary Figure S2 ). Overall, in E3FAD mice, OVX did not affect both learning/memory in MWM and Aβ pathology. E₂ improved only learning in MWM without impacting the Aβ levels/pathology.

Although E4FAD mice did not show any improvement in behavior but modulated Aβ pathology with E₂ treatment after early OVX, we evaluated the effect of OVX and subsequent E₂ treatment in older E4FAD mice ( Supplementary Figure S3 ). There was no treatment effect on latency to platform during MWM acquisition trials/probe trials, latency to target quadrant during probe trials, and insoluble Aβ42 and Aβ deposition in the cortex ( Supplementary Figures S3C–F ). Overall, we found that both OVX and E₂ treatment neither affected learning/memory nor Aβ pathology in E4FAD mice.

APOE3/3s account for ~30–50% of all AD patients (19). As female sex by itself is an AD risk factor, identifying pathways that could contribute to AD in APOE3/3 female individuals is important. One potential mechanism for higher AD risk is the loss of E₂ during and after the menopausal transition. However, clinical data on the impact of menopause on AD risk in APOE3/3s are limited as the focus is typically on APOE4. In APOE3-TR mice, OVX reduced hippocampal spine density, long-term potentiation (22), and disrupted learning in MWM (55). Our data extends those findings to APOE3-FAD mice, as we found that OVX disrupted memory in MWM. Therefore, the loss of sex hormones may be a major contributing factor to AD risk for a large proportion of patients. Based on that idea, E₂ would be predicted to protect against AD in APOE3 carrier. In fact, E₂ treatment was associated with less cognitive decline or higher learning and memory performance in post-menopausal APOE4 non-carriers (56, 57), and we found that E₂ improves memory in E3FAD mice (Early OVX). However, E₂ only had a modest effect on learning in late OVX in E3FAD mice. These are consistent with data that early oophorectomy, where there is likely low Aβ pathology that increases AD risk, is lowered by ERT (58). Although there are caveats, E₂ may be beneficial to prevent AD-associated neuronal dysfunction and cognitive decline in APOE3/3 if initiated early, before the accumulation of Aβ pathology.

Data from the current study and others raise the important discussion of the mechanism(s) that could underlie the effects of OVX and E₂ on neuron function and learning/memory. Our findings suggest that neither OVX nor E₂ modulate the Aβ levels in E3FAD mice, consistent with clinical data in APOE4 non-carriers (59). E₂ is a potent agonist of the transcriptional response of the nuclear hormone, estrogen receptors (ERα and ERβ), and also activates extranuclear ERx signaling (60). The beneficial effect of E₂ was likely mediated by ERs. ERs are expressed in multiple cell types throughout the brain (e.g., neurons, glia, and endothelial cells) and regulate signaling and gene expression that ultimately impact several functions—for instance, E₂ is thought to impact neuron function directly (61) and indirectly via effects on inflammation (62, 63), metabolism (64, 65), neurovascular function (66, 67), oxidative stress (68, 69), and APOE levels (61, 70). Indeed E₂ facilitated neurite outgrowth in APOE3 neurons (61) and suppressed inflammatory responses in APOE3 glia (71) in vitro. Linked to the question on how E₂ works is whether the different functions become disrupted with age and high Aβ pathology, which could result in a lower activity. Future studies could focus on identifying the critical functions of E₂ in APOE3/3 carriers.

AD risk is high in female APOE4 carriers (19–21), particularly at ages post-menopause. Therefore, it is logical to assume that APOE4 carriers should respond positively to E₂, yet clinical studies are more conflicted than for APOE3 (17, 72). The type, timing, duration, and dose of E₂ could contribute to discrepant results, along with APOE4-specific considerations. On the assumption that E₂ should be beneficial, the timing of treatment in relation to pathology may be critical for APOE4. In APOE4-TR mice, E₂ mitigated OVX-induced impairments in learning/memory (55). However, in the current study, E₂ was not beneficial in E4FAD mice. Female E4FAD mice have high levels of Aβ, AD-relevant pathologies, and memory deficits by 6–8 months of age (36, 73). Therefore, the combination of female sex, APOE4, and OVX may have resulted in a severe phenotype that was not recoverable by E_2. Thus, perhaps in less aggressive models that mimic gradual AD decline as found in humans, E₂ could have been beneficial. Further studies in E4FAD to determine whether E₂ by itself, administered earlier, may provide support to the “critical window” hypothesis.

An alternative explanation for the clinical data and our own is that, in the context of AD, APOE4/4s may be unresponsive to E₂. Many AD patients are APOE3/4, whereas here we focused on APOE4/4 (see limitations). There is evidence that APOE4/4 modulates E₂-dependent responses—for example, APOE4/4 neurons (61) and peritoneal macrophages from OVX APOE4-TR mice do not show a significant response to E₂ in vitro (71). The impact of APEO4 is pleotropic but includes altering receptor signaling, gene transcription, and lipid transport (74–76). Through those effects, APOE4 may blunt E₂-specific responses on multiple levels. In addition, the impact of OVX in the presence of APOE4 may also involve other sex steroid hormones and related hormones such as follicle-stimulating hormone (FSH)—for example, lowering FSH levels in APOE4/4 may mitigate the AD pathology and behavioral impairments associated with APOE4 (55). Thus, for APOE4/4s, it may be important to treat with other sex hormones such as progesterone as there is a link between APOE status and progesterone levels (77). Collectively, either initiating E₂ therapy earlier or treatment with other hormones may be beneficial for APOE4 carriers.

There are limitations in the extent to which we can conclude that E₂ can mitigate OVX-induced memory impairments in E3FAD mice. As discussed, it is important to conduct additional experiments to identify potential mechanisms through which E₂ induced a beneficial effect in E3FAD mice—for example, identifying whether these effects were mediated through ERα, Erβ, or ERx, cell-type specific effects, and downstream signaling pathways more proximal to behavior. Loss of circulating sex hormones in perimenopause is rapid within the context of a woman’s life, but rapid compared to surgical OVX, placing limitations on the model in both prevention and reversal paradigms (78). A less aggressive model of Aβ pathology than EFAD may also alter the response to E₂. The dose, frequency, drug formulation, and alternative estrogens may also influence response.

We are also limited to the extent that we can conclude E₂ is not beneficial for APOE4/4 carriers. As discussed, E4FAD mice are an aggressive model of Aβ pathology—for example, Aβ coverage in sham E4FAD mice is significantly greater than E3FAD mice that underwent sham surgery or OVX mice. Therefore, it is important to test the activity of E₂ in a less aggressive model or even in models without any mutations in APP/PSEN1. Related to this is that the 5xFAD genes are expressed via the Thy-1 promoter in EFAD mice, which is reported to contain an estrogen response element (ERE). However, it is likely that the base mutation (T/C→A) on the core consensus sequence of the Thy-1 promoter at the position +6 would abolish the ER–ERE interaction (79). In addition, the flanking sequence does not contain a purine at -7 position that could potentially increase the binding affinity (80, 81). Furthermore, if E₂ did bind to the ERE, then it would occur equally in E3FAD and E4FAD and would unlikely explain the APOE genotype differences in response in this study. However, further studies in additional APOE4 models are important. There were also some experimental limitations surrounding the way we induced hormonal loss. One aspect is how the extent of sex hormone loss induction (either OVX or chemical) in mice translates to humans is unclear with APOE4, especially as female E4FAD mice have disrupted behavior in the absence of OVX (36, 73). In fact, evaluating the activity of E₂ in E4FAD mice or other mouse models in the absence of menopausal mimic may provide information of how E₂ can impact brain function during aging. There is also the question of whether there are differences in cell-type-specific distribution of ERs, downstream signaling molecules (see “Discussion”), and functions in E3FAD compared to E4FAD mice after OVX. Therefore, conducting additional experiments is critical before discounting the potential of E₂ as a therapeutic target for preventing/treating AD in female APOE4/4s individuals.

General limitations also include a lack of pharmacokinetic studies to determine brain and plasma levels after treatment. Although the effects of treatment are evident in estrogen-sensitive gynecological tissues, the sensitivity of tissues to estrogens may not reflect menopause. We selected the dose of E₂ based on previous publications; however, levels in the plasma and brain may have been sub-optimal or even different between mice. In addition, continuous delivery of estrogen results in sustained estrogen levels (82–85), which does not mimic physiological fluctuations in hormonal levels, and therefore cyclic E₂ treatments may provide greater neural protection. However, the efficacy of cyclic vs. continuous delivery of estrogen is controversial with studies showing improvement in learning/memory using continuous treatment (86–89), cyclic treatment (90), and continuous treatment when primed with repeated injections of E₂ (91).

Another limitation of this study is that we are unable to directly compare data obtained in E3FAD and E4FAD mice, as the analysis of each APOE genotype was conducted separately due to the COVID-19 pandemic. In addition, it is also important to incorporate APOE3/4, additional hormones, and treatment windows in future studies.

Our data supports that the APOE differentially modulated the effect of OVX and E₂ on behavior and Aβ pathology—specifically, that E₂ may benefit APOE3/3s but not APOE4/4s after loss of sex hormones. Future studies are critical to the optimal treatment approaches for addressing the increased risk of AD after menopause for each APOE genotype.