Effects of early midlife ovarian removal on sleep: Polysomnography-measured cortical arousal, homeostatic drive, and spindle characteristics

, individuals with BSO show reduced hippocampal volume, function, and hippocampal-dependent verbal episodic memory performance associated with changes in sleep. It is unknown whether BSO affects fine-grained sleep measurements ( sleep microarchitecture ) and how these changes might relate to hippocampal-dependent memory. We recruited thirty-six early midlife participants with BSO. Seventeen of these participants were taking 17 β -estradiol therapy (BSO + ET) and 19 had never taken ET (BSO). Twenty age-matched control participants with intact ovaries (AMC) were also included. Overnight at-home polysomnography recordings were collected, along with subjective sleep quality and hot flash frequency. Multivariate Partial Least Squares (PLS) analysis was used to assess how sleep varied between groups. Compared to AMC, BSO without ET was associated with significantly decreased time spent in non-rapid eye movement (NREM) stage 2 sleep as well as increased NREM stage 2 and 3 beta power, NREM stage 2 delta power, and spindle power and maximum amplitude. Increased spindle maximum amplitude was negatively correlated with verbal episodic memory performance. Decreased sleep latency, increased sleep efficiency, and increased time spent in rapid eye movement sleep were observed for BSO + ET. Findings suggest there is an association between ovarian hormone loss and sleep microarchitecture, which may contribute to poorer cognitive outcomes and be ameliorated by ET.


Introduction
Early midlife bilateral salpingo-oophorectomy (BSO; removal of ovaries and fallopian tubes) is associated with accelerated cognitive decline and increased Alzheimer's disease (AD) risk (Bove et al., 2014;Rocca et al., 2007).Even within five years post-BSO, participants show cognitive and brain changes possibly presaging late-life AD (Brown et al., 2023;Gervais et al., 2022Gervais et al., , 2020)).Women with early midlife ovarian removal have polysomnography-measured sleep disturbance comparable to older women in spontaneous menopause (Gervais et al., 2023).They also report increased sleep disturbance and have a higher prevalence of sleep disorders than age-matched participants in spontaneous menopause, which is especially important considering that the midlife experiences of sleep disorders like insomnia and sleep apnea are associated with increased dementia risk (Baker et al., 2018;Cho et al., 2019;Lim et al., 2013;Przybylska-Kuć et al., 2019;Sindi et al., 2018;Xu and Lang, 2014).Despite the high prevalence and increased severity of sleep disturbance characterizing AD, the extent to which early midlife ovarian hormone deprivation contributes to sleep disturbance and AD risk is understudied.Particularly, no research has focused on whether the fine-grained phasic events characterizing electroencephalography (EEG)-measured sleep microarchitecture are affected by BSO.This is an important research gap given that microarchitecture measures may offer more precise insights into underlying neurobiological mechanisms implicated in the development and progression of dementia compared to the broader assessments of overall sleep patterns measured by traditional sleep macroarchitecture, which may underestimate the extent of sleep disturbance when brain changes are subtle (Schwarz et al., 2017).
Given that sleep microarchitecture might provide more information about the neurobiological mechanisms affected by BSO, the primary objective of this study was to investigate sleep microarchitecture in this younger midlife cohort.While our previous work has focused on sleep macroarchitecture, we hypothesize that sleep microarchitecture might be particularly sensitive to detecting subtle changes preceding more obvious signs of dementia risk.Further, no studies have investigated potential links between sleep microarchitecture and cognition in younger women with BSO.We sought to fill this research gap by identifying some of the earliest and most subtle sleep changes associated with dementia risk, which might help to improve understanding of any underlying neural changes affecting sleep in women with BSO.

Sleep macroarchitecture and BSO
Work from our lab suggests BSO without 17β-estradiol therapy (ET) is associated with changes in sleep macroarchitecture, including increased time taken to fall asleep and less time spent asleep in bed (Gervais et al., 2023).This greater time taken to fall asleep is related to lower verbal episodic memory performance and smaller anterolateral cortex volume (Gervais et al., 2023).To date, no known research has investigated the intersection of ovarian hormones, sleep macroarchitecture, sleep microarchitecture, and memory, particularly within the context of early midlife BSO.Acknowledging that macro-and microarchitecture are not isolated from one another, these factors should be studied together following early midlife BSO to best grasp how these interrelated variables are affected by ovarian hormone loss.We posit that considering macro-and microarchitecture variables simultaneously-compared to macroarchitecture alone-will provide a more detailed understanding of the underlying sleep mechanisms affected by BSO and how affected circuits could affect awake cognition.

Sleep microarchitecture and BSO
Despite the increased insomnia risk related to BSO, it is unknown whether BSO is associated with insomnia-related sleep microarchitecture patterns, such as increased power in the beta frequency range (15-30 Hz), reflecting cortical hyperarousal and compromised sleep quality (Buysse et al., 2008;Chappel-Farley et al., 2020;Perlis et al., 2001;Rezaei et al., 2019;Stone et al., 2008).Non-rapid eye movement (NREM) beta power, sleep complaints, and insomnia risk increase across the spontaneous menopause transition, suggesting microarchitectural cortical hyperarousal may be linked to ovarian hormone loss (Campbell et al., 2011;Matthews et al., 2021).Five years prior to mild cognitive impairment or dementia diagnosis, older women are also more likely to show cortical hyperarousal during sleep compared to those who remain cognitively healthy (Djonlagic et al., 2019).Results from these studies cumulatively suggest ovarian hormone loss affects cortical hyperarousal during sleep, which may exacerbate dementia risk.
Power in the delta frequency range (0.3-4 Hz), which characterizes homeostatic sleep drive and deep/restorative slow-wave sleep, has not been studied in women with BSO.Results of research focused on delta power across the spontaneous menopausal transition are inconsistent, with some work demonstrating that delta power decreases (Kalleinen et al., 2008) or is unaffected (Campbell et al., 2011).These results are surprising given that spontaneous menopause is associated with increased time spent in NREM stage 3 slow-wave sleep (Hachul et al., 2015;Sowers et al., 2008).Thus, while spontaneously menopausal women may spend more time in deep NREM stage 3, this may not reflect more intense sleep.Importantly, abnormalities in delta power are common in AD.For example, delta power is increased with amyloid β burden and mild AD (Babiloni et al., 2013;Katsuki et al., 2022).Thus, delta power may provide information about the onset and progression of important pathological AD-related changes.

Sleep spindles and AD risk
Sleep microarchitectural spindles, characterizing NREM stage 2 and generated by gamma-aminobutyric acid (GABA)-ergic neurons of thalamic nuclei, are short neural bursts of variable peak amplitude (~100 μV) and duration (0.5-2 s) within the sigma frequency range (11-16 Hz) (Bandarabadi et al., 2020;Timofeev and Chauvette, 2013).Spindles are thought to stabilize sleep, gate sensory processing, and coordinate neural activity between the hippocampus and cortex to consolidate memory (Cowan et al., 2020;Dang-Vu et al., 2011).Based on frequency and topography, spindles can be categorized into frontal slow spindles (11-13.5 Hz) and centro-parietal fast spindles (13.5-16Hz) (Mölle et al., 2011).Both slow and fast spindles support hippocampaldependent memory consolidation in men and women (Lustenberger et al., 2015;Schabus et al., 2008).These spindles reach the medial temporal lobe by either direct projection from the thalamus to the posterior hippocampus (Dolleman- Van der Weel et al., 1997;Poppenk and Moscovitch, 2011) or the neocortex to the entorhinal cortex (Isomura et al., 2006).When a spindle successfully enters this medial temporal lobe network, it can trigger a hippocampal sharp-wave ripple, initiating neuronal replay and synchronizing activity across different cortical areas.Thus, spindles may be viewed as indicators of the integrity of the memory consolidation process and serve as crucial indicators of AD progression and decline in cognitive function (Cowan et al., 2020).
Women tend to show more variability in spindle characteristics compared to men (Martin et al., 2013;Purcell et al., 2017;Weng et al., 2020).This spindle variability may be related to ovarian hormone level fluctuation.The number of spindles and power in the spindle sigma frequency range both increase during the luteal phase of the menstrual cycle, when 17β-estradiol levels are low and progesterone levels are moderate (Baker and Lee, 2022;Ishizuka et al., 1994).Research has begun to investigate the implications for this spindle variability, with findings suggesting spindles mediate sleep-dependent memory consolidation in a menstrual cycle phase-specific manner (Sattari et al., 2017).However, variations in spindles and their influence on hippocampaldependent memory among women with BSO have not yet been studied.

Sleep disturbance and the hippocampus
Brain structures affected by early midlife BSO and disturbed sleep, including the hippocampus, coincide with brain structures first affected in AD.Sleep plays a crucial role in the consolidation of hippocampaldependent memories, suggesting the hippocampus may be particularly sensitive to the consequences of sleep disturbance (Kreutzmann et al., 2015).During NREM, the hippocampus is critical for consolidation of episodic memory via integration of memory representations across hippocampal and cortical regions (Ji and Wilson, 2007;Lee and Wilson, 2002;Louie and Wilson, 2001;Nádasdy et al., 1999).Mixed sex studies show decreased hippocampal volume is related to poor subjective sleep quality (Fjell et al., 2020;Liu et al., 2021), increased frequency of arousals during sleep, as well as increased insomnia diagnosis (Joo et al., 2014) and diagnosis duration (Noh et al., 2012).Women with early midlife BSO without ET have reduced posterior hippocampal activation (Brown et al., 2023) and volume loss specific to the hippocampal dentate gyrus cornu ammonis 2/3 composite region (Gervais et al., 2022).Taken together, these findings suggest that subtle midlife hippocampal changes in women with BSO may relate to sleep disturbance.Given that the hippocampus is susceptible to the negative effects of insufficient sleep, it is possible that early loss of ovarian hormones negatively affects sleep, resulting in hippocampal atrophy (Prince and Abel, 2013).

Study objectives
Considering significant gaps in the understanding of effects of early midlife BSO on sleep microarchitecture and its correlation with sleep macroarchitecture and hippocampal-dependent memory, we used athome polysomnography (PSG) to compare sleep architecture between early midlife participants with BSO who were not taking ET (BSO), those taking ET (BSO+ET), and age-matched control participants with intact ovaries (AMC).We predicted that without ET, participants with BSO would have: 1) Decreased sleep quality measured by macroarchitecture, 2) Cortical hyperarousal (increased NREM beta power), 3) Increased homeostatic sleep drive (increased NREM delta power), 4) Alterations in spindle characteristics, and 5) The effect of BSO on sleep microarchitecture would have negative implications for hippocampaldependent verbal episodic memory performance.

Participant recruitment
This study was carried out in accordance with the principles of the Declaration of Helsinki.Approval was granted by the Research Ethics Committees of the University of Toronto and McGill University.Informed consent was obtained from all participants.
Participants with BSO were recruited from familial breast and ovarian cancer clinics in Toronto and Montreal, Canada.AMC participants were recruited from the general community in the same cities.Exclusion criteria for all participant groups included: being younger than 35 or older than 55 years, perimenopause, BSO after spontaneous menopause, pregnancy, breastfeeding, chemo/radiation/adjuvant therapies within six months of testing, and unmanaged health/psychiatric conditions and/or endocrine disorders.There were two additional exclusion criteria for AMC: irregular menstruation and taking hormonal contraceptives within six months prior to entering the study.Participant demographic characteristics are summarized in Table 1.

Procedure
Data included in this study are a subset of a larger longitudinal dataset addressing neuropsychological performance, brain structure, brain function, and sleep of women with BSO (Brown et al., 2023;Gervais et al., 2023Gervais et al., , 2022Gervais et al., , 2020)).During a two-hour session, participants were administered demographic, mood, and sleep questionnaires as well as various cognitive tests.Mood measures included the Centre for Epidemiological Studies-Depression scale (CES-D) (Radloff, 1977), which assessed depressive mood, and the Perceived Stress Scale (PSS) (Cohen et al., 1983), which assessed stress symptoms.The Pittsburgh Sleep Quality Index (PSQI) questionnaire was administered to assess sleep quality over the previous one-month time interval (Buysse et al., 1989).Participants also completed a sleep diary following each night of recorded sleep, as described elsewhere (Gervais et al., 2023).Neuropsychological task and questionnaire data were not always collected at the same session as the sleep data.There was an average delay of 142 days between collection of neuropsychological and sleep measures, with most sessions occurring within three months of each other.
During their testing session, participants also provided urine samples for ovarian hormone level assessment.Levels of estrone-3-glucuronide (E1G) and pregnanediol glucuronide (PdG), the main secreted forms of circulating 17β-estradiol and progesterone in mammals, respectively, were analyzed in urine using enzyme-linked immunoassays at the Women's Health and Exercise Laboratory at Pennsylvania State University as described elsewhere (Munro et al., 1991).These immunoassays could detect 17β-estradiol and progesterone contained in hormone therapy but did not discriminate between exogenous and endogenous forms of these hormones.Although urine collection at the same time of day for all participants would have been ideal to account for potential circadian fluctuations in hormone metabolite excretion (Rahman et al., 2019), for convenience time of collection was adjusted to the timing of the testing session and varied between participants.
Overnight PSG recordings were collected by the participants at their homes using a portable research-grade PSG device (Vitaport-5/REMbo-234, Type-II, Temec Technologies, The Netherlands).Following detailed instructions and practice with an experimenter, participants were given thorough written and video step-by-step instructions for applying the device at home.Experimenters contacted participants following the first night of sleep measurement to ensure that PSG application had gone smoothly.Participants were also instructed to contact the experimenter if they had issues with PSG application.All participants were instructed to record their sleep for three nights, however scheduling challenges and occasional discomfort led to some participants recording fewer nights of sleep; therefore, the number of recordings per participant varied between one and three nights (M = 2.21, SD = 0.80).

Verbal episodic memory
We assessed hippocampal-dependent verbal episodic memory using a paragraph recall task (Logical Memory, Wechsler Memory Scale Form I) (Frisk and Milner, 1990;Wechsler, 1945).Participants were read a  0 (0)  1 (5.9)  9 (47 A. Brown et al. brief story before recalling details both immediately (immediate trial) and approximately 30 min after presentation (delayed trial).Correctly recalled verbatim details were summed, providing scores for immediate and delayed trials (Abikoff et al., 1987).Task scores were missing for two participants from the AMC group who were not naïve to the task purpose and story details; therefore, the task was not administered to these participants and their data were excluded from analyses involving this measure.This verbal episodic memory task was selected because past work from our lab showed reduced performance with early midlife BSO (Gervais et al., 2020) significantly related to increased sleep latency (Gervais et al., 2023).We focused mainly on task performance during the delayed trial because it is thought to be particularly dependent on the hippocampus (Frisk and Milner, 1990).

Polysomnography recordings
As described elsewhere (Gervais et al., 2023), sleep recordings included signals from various wet electrodes (Ambu Disposable Blue Sensor M), including two frontopolar (Fp1/Fp2) EEG electrodes, and two electrooculography (EOG) electrodes (LOC/ROC), referenced to the left mastoid.An additional signal estimating facial electromyography from the two EEG and two EOG electrodes was also included.These signals were digitized at 256 Hz, with a filter of 60 Hz.Because PSG directly measures brain electrophysiology, it is often considered the "gold standard" measure of sleep (Lehrer et al., 2022).At-home PSG systems, including the one from Temec, enhance ecological validity, have similar accuracy to laboratory PSG for sleep disorder diagnosis (Lachance and Bailey, 2023), and have been previously used to assess sleep architecture across the menopause transition (Campbell et al., 2011;Sowers et al., 2008).For sleep architecture, single-channel EEG also has strong agreement with full-montage PSG, suggesting Fp channels alone can provide sufficient information (Lucey et al., 2016).Further, other studies have similarly focused on frontal derivations to measure sleep microarchitecture, including spindles (Sattari et al., 2017).
All recordings were viewed in 30-s epochs.The sleep stage of each epoch (Wake, NREM stage 1 (N1), NREM stage 2 (N2), NREM stage 3 (N3), and Rapid Eye Movement (REM)) was automatically scored following a modified version of the American Academy of Sleep Medicine criteria to account for two frontopolar electrodes using Z3score (version 2.2.0, Neurobit Technologies) (Berry et al., 2017).This automatic method has strong sensitivity and specificity for staging sleep recordings.Additionally, the staging interrater reliability between this method and human experts is similar to the interrater reliability found between human experts (Choo et al., 2023;Goldstein et al., 2020;Patanaik et al., 2018).To ensure accurate staging, all automatically scored recordings were also visually inspected by trained experimenters blind to group membership.Experimenters determined an overall quality rating (0-3) based on the availability and integrity of each EEG/ EOG signal.Recordings were included in analyses if at least one EEG channel of good quality was available (recording quality rating ≥ 1.5).In total, 44/174 recordings were excluded due to poor recording quality and excessive artifact (n = 30 participants).Of the 30 participants with poor quality recordings, 21 participants had at least one other usable/ good-quality recording and nine participants were fully excluded because they had no useable recordings.Experimenters manually adjusted automatic sleep staging for all recordings if the selected stage did not align with the American Academy of Sleep Medicine criteria.For microarchitecture analyses, trained experimenters also manually scored arousals and signal artefacts for each recording.All sleep parameters were averaged across each participant's recordings.Sleep macro-and microarchitecture data were extracted from each recording using the Wonambi package for python (https://wonambi-python.github.io/).

Sleep microarchitecture 2.5.1. Spindle detection
Sleep spindles were automatically detected using Wonambi (http s://wonambi-python.github.io/)with the algorithm from (Lacourse et al., 2019).Given that our EEG configuration included frontopolar electrodes, we focused on data for spindles thought to be maximally expressed in the frontopolar region, those within the slow rather than the fast frequency range (Mölle et al., 2011).Spindles with a peak frequency between 11 and 13.5 Hz were classified as slow spindles (Mednick et al., 2013).Both artefacts and arousals were excluded from the signal for spindle analyses.During N2 and N3, spindle measures of interest included spindle density (number of spindles per minute), spindle duration (seconds), and spindle maximum amplitude (μV).

Spectral power
Spectral power data were also extracted from each recording using Wonambi (https://wonambi-python.github.io/).Spectral power analysis measures the intensity of a time-varying signal and how it is distributed over a frequency band to better understand the properties and magnitude of that signal.We conducted band-limited spectral power analysis within the beta (15-30 Hz), delta (0.3-4 Hz), and slow spindle (11-13.5 Hz) frequency ranges for each frontopolar channel (Fp1 and Fp2), calculating absolute power across the two channels.If only one channel had been recorded, power was calculated for that channel only.Artefacts were removed from the signal and the resulting signal segments belonging to a same sleep stage (N2, N3) were concatenated.Segments shorter than 15 s were discarded.Welch's method (1-s Hann window, 50 % overlap) was then applied to each resulting segment (participant x sleep stage) to obtain the average spectral power density.Band-limited EEG power was then averaged separately for N2 and N3 stages.To achieve normal distributions, absolute power data were then log-transformed.During N2 and N3, spectral power measures of interest included N2 and N3 log-transformed beta, delta, and slow spindle power (log 10 μV 2 ).

Statistical analyses
All statistical analyses were conducted using R 4.0.2(R Core Team, 2024).Statistical assumptions were tested, including normality and homogeneity of variance.When violated, we Winsorized data to the value of the 90th or 10th percentile of the distribution.To compare groups on participant demographics and subjective sleep during the study, analysis of variance (ANOVA) or Fisher's exact tests were used including group as a between-participants factor.In case of a significant main effect of group, multiple comparison corrected Tukey post hoc analyses were carried out to disentangle the effect.If model assumptions were violated, Mann-Whitney U tests or Kruskal-Wallis tests followed by post-hoc Dunn's tests were conducted.Effect size estimates (η 2 or Cohen's d) were calculated for all parametric analyses.

Partial least squares correlation analysis: how does BSO affect sleep macro-and microarchitecture?
To determine how BSO affects sleep macro-and microarchitecture, we conducted a Partial Least Squares (PLS) correlation analysis in R using data4PCCAR and TExPosition.This multivariate approach allowed us to model the relationship between group status (BSO, BSO+ET, AMC) and sleep architecture.Macroarchitecture variables included sleep latency, sleep efficiency, and percentage of total sleep time (TST) spent in each of the three stages of NREM (N1− N3) and REM sleep.Microarchitecture variables included N2 and N3 log-transformed beta, delta, and spindle power (log 10 μV 2 ), spindle density (number of spindles per minute), spindle duration (seconds), and spindle maximum amplitude (μV).Due to skewness and to best approximate normal distributions, data for sleep latency, sleep efficiency, and percentage of TST spent in N3 were Winsorized (to the value of the 90th or 10th percentile of the distribution) and data for beta, delta, and spindle power were logtransformed.Data were centered and standardized prior to analysis using the preProcess function from caret.
Next, data were organized into two matrices, one of group status, and another of the sleep architecture variables.PLS was used to model the relationships between the variables in these two matrices.The resulting matrix was then subjected to singular value decomposition, a data reduction technique that circumvents the need for multiple comparison corrections because it is conducted in one step.Consequently, compared to univariate analysis, multivariate PLS offers heightened sensitivity and robustness and is a good approach for handling datasets with smaller cohort sizes (Lukic et al., 2002;Willaby et al., 2015).Unlike univariate analyses, PLS does not require meeting assumptions of normality, independence of observations, and linearity, and allows all variables to be modelled simultaneously (Van Roon et al., 2014).Two resampling methods were used to identify which latent variables were significant (permutation tests with 1000 iterations), and which variables were the most reliable contributors to each latent variable (1000 bootstrap samples) (Beaton et al., 2014;Berry et al., 2011;Efron and Tibshirani, 1986;Hesterberg, 2011;Peres-Neto et al., 2005).
We also ran a separate exploratory supplementary PLS analysis modelling the relationship between group status (BSO, BSO+ET, AMC), the presence or absence of self-reported nocturnal hot flashes, and sleep architecture (Supplementary Material).

Pearson correlation
Pearson correlation was used to understand whether group differences in sleep microarchitecture related to performance on the verbal episodic memory task by examining the relationship between verbal episodic memory (delayed paragraph recall performance) and the sleep microarchitecture variable showing the largest PLS effect.

Demographic characteristics
As expected, BSO, BSO+ET, and AMC did not differ significantly in age at study entrance (F(2,53) = 1.36, p = 0.27, η 2 = 0.05).BSO and BSO+ET also did not differ in age of BSO (t(34) = 0.66, p = 0.52, d = 0.22) or time since BSO (t(34) = 0.58, p = 0.56, d = 0.20).There was a significant effect of group on past cancer treatment history (chemotherapy, radiation therapy, and/or adjuvant therapy; p < 0.0001), with more women in the BSO group having a history of cancer treatment compared to those in the BSO+ET (p = 0.01) and AMC (p = 0.001) groups.
Data regarding urinary ovarian hormone levels were not available for a small subset of participants (E1G: BSO n = 4, BSO+ET n = 4, AMC n = 3; PdG: BSO n = 4, BSO+ET n = 5, AMC n = 3).For participants for whom urinary ovarian hormone level data were available, there was a significant effect of group on urinary E1G levels (χ 2 = 17.67, p = 0.0001): BSO (Z = 4.09, p = 0.0001) and BSO+ET (Z = 2.69, p = 0.01) had significantly lower E1G levels than AMC.E1G levels did not significantly differ between BSO and BSO+ET (Z = − 1.21, p = 0.23).There was also a significant effect of group on urinary PdG levels (χ 2 = 6.35, p = 0.04): BSO trended toward having significantly lower PdG levels than BSO+ET (Z = − 1.96, p = 0.07) and AMC (Z = 2.33, p = 0.06).PdG levels did not significantly differ between BSO+ET and AMC (Z = 0.18, p = 0.86).The lack of significant group differences in urinary PdG levels could have been due to variability in menstrual phase and progesterone therapy use for the AMC and BSO+ET groups, respectively.For example, those in the BSO+ET group who were taking progesterone therapy had significantly higher urinary PdG levels than those in the BSO+ET group who were not taking progesterone therapy (Z = − 2.36, p = 0.02).For the BSO+ET group, the average duration of taking ET was 3.27 years (ranging from two months to 12 years).Hormone therapy details are summarized in the Supplementary Material.There were no significant group differences for any other demographic measures (Table 1).

Subjective and objective sleep
Self-reported sleep data (Table 2) and objective PSG macro-and microarchitecture sleep data (Table 3) are presented to provide general information about sleep quality between groups.Sleep medication details are also summarized in Table 4.There were no significant group differences for most subjective sleep measures, including the number of participants in each group taking sleep medication.Further, neither the number nor the quality of PSG recordings differed between groups.There was a significant effect of group on perceived sleep quality during the previous month as measured by the PSQI (χ 2 = 6.61, p = 0.04); BSO had significantly higher scores (indicating worse sleep quality) than AMC (Z = − 2.46, p = 0.04), but there were no significant differences between BSO and BSO+ET (Z = 0.55, p = 0.58) and between BSO+ET and AMC (Z = − 1.92, p = 0.08).There was also a significant effect of group on the number of self-reported nocturnal hot flashes per night (χ 2 = 15.19,p = 0.001); BSO reported significantly more hot flashes than AMC (Z = − 3.90, p = 0.0003), but there were no significant differences between BSO and BSO+ET (Z = 2.05, p = 0.06) and between BSO+ET and AMC (Z = − 1.74, p = 0.08).
LV1 indicated that BSO without ET was associated with significantly decreased time spent in N2 sleep (% TST) as well as increased N2 and N3 beta power, N2 delta power, and spindle power and maximum amplitude.AMC was associated with significantly increased time spent in N2 sleep as well as decreased N2 and N3 beta power, N2 delta power, and spindle power and maximum amplitude (Fig. 1A-B).For LV1, BSO+ET was not significantly related to any macro-or microarchitecture variables.Bootstrap ratios for LV1 are listed in Supplementary Material Table S4.
LV2 indicated that BSO+ET was associated with significantly decreased sleep latency, as well as increased sleep efficiency and time spent in REM sleep compared to BSO and AMC (Fig. 2A-B).Bootstrap ratios for LV2 are listed in Supplementary Material Table S5.

Relationship between slow spindle maximum amplitude and verbal episodic memory
There was not a significant effect of group status on performance (number of details recalled) for the immediate trial of the verbal episodic memory paragraph recall task (M = 6.02,SD = 2.64, F(2,51) = 0.24, p = 0.79, η 2 = 0.01).There was a trend toward a significant effect of group status on performance (number of details recalled) for the delayed trial of the verbal episodic memory paragraph recall task (M = 4.06, SD = 2.32, F(2,51) = 2.75, p = 0.07, η 2 = 0.10), with BSO trending toward performing significantly worse than AMC (t(51) = 2.24, p = 0.07, d = 0.74), and no significant differences between BSO and BSO-+ET (t(51) = − 1.67, p = 0.23, d = − 0.56) or between BSO+ET and AMC (t(51) = 0.53, p = 0.86, d = 0.18).Because the effect of group status on task performance was not statistically significant, data was pooled across groups for correlational analyses.Additionally, exploratory analysis revealed that there was not a significant interactive effect of group status and spindle maximum amplitude on delayed trial performance (F(2,48) = 0.80, p = 0.46, η 2 = 0.03).This result suggests that the relationship between spindle maximum amplitude and verbal episodic memory performance did not depend on group status.Thus, regardless of group status, greater slow spindle maximum amplitude was associated with poorer performance on the delayed trial of the verbal episodic memory paragraph recall task (r(52) = − 0.29, p = 0.03; Fig. 3).

Summary of findings
This is the first study of sleep microarchitecture in early midlife women with BSO as well as the first to correlate sleep microarchitecture with memory.We found that women with BSO without ET spent less time in N2 and had higher N2 and N3 beta power, N2 delta power, and spindle power and maximum amplitude compared to AMC.BSO+ET was not significantly different than BSO or AMC, suggesting ET may provide some, but not complete, protection against this pattern of sleep effects.Aligning with our previous work in the same cohort (Gervais et al., 2023), we also found that ET may protect against important sleep macroarchitecture patterns, including decreased sleep efficiency and increased sleep latency.Taken together, these studies offer a comprehensive view of macro-and microarchitecture sleep patterns after early midlife ovarian removal.

BSO without ET was associated with decreased time spent in N2 sleep
This is the first study to show that compared to AMC, women with BSO without ET spent less time in N2.Past work has demonstrated that for men and women, aging is associated with less time spent in N2, although men tend to spend more time in N2 than women overall (Purcell et al., 2017).Thus, it is possible that with respect to sleep, women with BSO show an accelerated aging brain phenotype, consistent with other work linking BSO to accelerated aging of multiple bodily systems (Rocca et al., 2017).
Although our past work did not find that BSO affected time spent in N2 (Gervais et al., 2023), the current study used multivariate PLS to analyze numerous variables simultaneously, acknowledging the interrelated aspects of sleep macro-and microarchitecture.PLS findings allow us to consider that without ET following BSO, decreased time spent in N2 might occur concomitant to an increase in the sleep spindle activity that characterizes this sleep stage and may point to a reason why spindle activity did not significantly correlate with time spent in other sleep stages with less frequent spindle events (REM, N3).Though speculative, women with BSO may offset reduced time spent in N2 sleep by augmenting spindle amplitude.This potential compensatory mechanism may serve as an adaptive response to early ovarian hormone loss.Future research should elucidate and confirm these relationships and determine their underlying neurobiological mechanisms.

ET was associated with improvements in sleep macroarchitecture
We found that compared to BSO and AMC groups, women with BSO who were taking ET had increased sleep efficiency, decreased sleep latency, and spent the most time in REM sleep.Results are consistent with work from our lab showing that BSO without ET is associated with decreased sleep efficiency and increased sleep latency (Gervais et al., 2023).They also align with studies showing that spontaneously menopausal women taking ET tend to have improved sleep quality, shortened sleep latency, and increased time spent in REM compared to those without ET (Antonijevic et al., 2000;Kravitz et al., 2008;Polo-Kantola et al., 1998).Some of the strongest support for the positive effects of hormone therapy on sleep comes from a study of older spontaneously menopausal women (57-80 years) without hot flashes who were taking either conjugated equine estrogen or esterified estrogens (estrone/ Equilin).These women experienced nighttime blood draws and their sleep was compared with those who were not taking hormone therapy.The hormone therapy group had less disrupted sleep as evidenced by reduced sleep latency, time awake, and increased sleep efficiency (Moe et al., 2001).Together, these studies support the subtle benefits to sleep macroarchitecture of ET.Abbreviations: AMC = age-matched controls, BSO+ET = bilateral salpingooophorectomy with current use of 17β-estradiol therapy, BSO = bilateral salpingo-oophorectomy.

BSO without ET was associated with cortical hyperarousal
Consistent with their increased insomnia risk (Cho et al., 2019), women with BSO without ET had heightened beta power, associated with cortical hyperarousal.Cortical hyperarousal within the beta frequency range indicates less restful sleep and in mixed-sex studies has been associated with aging (Carrier et al., 2001).Increased beta power has also been noted among perimenopausal women with insomnia (Baker et al., 2015), and across the spontaneous menopause transition, even after controlling for age (Campbell et al., 2011).These studies support our findings that even in the absence of aging as a confound between groups, ovarian hormone loss alone may be associated with less restful sleep.Supporting this are the associations of shifts in 17β-estradiol and follicle stimulating hormone levels during perimenopause with poorer sleep quality and more awakenings (Dennerstein et al., 2007;Kravitz et al., 2008;Sowers et al., 2008).Future work should focus more particularly on how fluctuating ovarian hormone levels affect beta power and dementia risk among women with BSO.
Nocturnal hot flashes may also contribute to cortical hyperarousal.Hot flashes are reported by 60 % to 90 % of spontaneously peri/postmenopausal women (Gold et al., 2006;Williams et al., 2008;Woods and Mitchell, 2005), with frequency correlating with greater sleep complaints and reduced sleep efficiency (De Zambotti et al., 2014;Freeman et al., 2015).Compared to spontaneous menopause, hot flashes are observed in greater numbers and with increased severity in women with early oophorectomy (Kingsberg et al., 2020).In our study, the number of hot flashes reported by women was significantly higher in BSO compared to AMC but did not differ between BSO and BSO + ET.When including the presence of reported hot flashes as a variable in the PLS analysis (Supplementary Material), their effect on sleep architecture was not significant, suggesting hot flashes were likely not the sole cause of cortical hyperarousal in our cohort.Other studies have shown similar patterns.One showed fragmented sleep increased across the spontaneous menopausal transition, even after statistically controlling for selfreported hot flashes (Lampio et al., 2017).Still, women tend to underreport the number of hot flashes that they experience, so it will be important for future studies to measure objective hot flashes to best tease apart the effects of ovarian hormones and vasomotor symptoms on sleep (Maki et al., 2008).Our findings emphasize that sleep disturbance, including cortical hyperarousal, among women with BSO may not be completely explained by self-reported nocturnal hot flashes and likely has a multifactorial basis that must be treated by an integrated and cross-disciplinary approach to sleep care (Baker, 2023).

BSO without ET was associated with increased homeostatic drive
This is the first study to show that, without ET, women with BSO exhibited heightened delta power.Increased delta power might reflect more intense/deep sleep perhaps as a recovery response to reduced sleep quality.These findings align with previous research indicating that time spent in N3 slow-wave sleep increases across the spontaneous menopause transition (Hachul et al., 2015).Given that time spent in N3 slowwave sleep did not vary significantly between groups in our study, delta power may be a more sensitive metric for assessing homeostatic sleep drive.Notably, our results demonstrate that the BSO group did not have significantly increased delta power when accounting for self-reported hot flashes in the PLS model, suggesting this effect may be partially explained by the experience of hot flashes (Supplementary Material).This idea is consistent with work showing that slow-wave sleep is facilitated by increasing brain temperatures (McGinty and Szymusiak, 1990).Additionally, research indicates that for spontaneously menopausal women, hot flashes occurring in the two hours prior to sleep onset positively correlate with the amount of slow-wave sleep (Woodward and Freedman, 1994).Furthermore, in women with breast cancer, hot flash A. Brown et al. frequency is significantly correlated with increased delta power (Savard et al., 2013).Together, these findings highlight the complex relationship between sleep architecture and physiological changes related to ovarian hormone milieu.

BSO without ET was associated with increased spindle power and amplitude
Consistent with the idea that ovarian hormone levels influence spindle characteristics, we found women with BSO without ET had significantly increased spindle power and maximum amplitude compared to AMC, while BSO + ET did not differ from either group.
Other work in women has demonstrated increased spindle frequency (Hz) during the late luteal menstrual phase, characterized by low 17βestradiol and elevated progesterone levels (Driver et al., 1996;Ishizuka et al., 1994).Thus, progesterone may also influence spindles (Plante and Goldstein, 2013).As 17β-estradiol levels increase across pregnancy, there is also a reduction in spectral power within the spindle frequency range, aligning with an inverse relationship between 17β-estradiol levels and spindles (Brunner et al., 1994).Further, most mixed-sex studies report a progressive decline in the number of spindles per night of sleep with increasing age (Carrier et al., 2001;Guazzelli et al., 1986;Martin et al., 2013;Nicolas et al., 2001).These changes may be attenuated in women (Martin et al., 2013).Thus, even though the number of spindles may decline with age on average in women, they may also be experiencing more variability in this pattern due to ovarian hormone milieu.
Increased spindle power and amplitude among women with early midlife BSO is unexpected in the context of their increased risk of latelife AD given that mixed-sex studies have shown decreased spindle amplitude associated with AD progression (Liu et al., 2020).However, most research has focused on elderly cohorts of men and women while statistically controlling for sex, with little to no research concentrated on the spindle characteristics of women in midlife.Therefore, it is unclear whether and at which point spindle activity declines might begin to occur in this population.
Though speculative, it is interesting to consider that increased spindle activity may compensate for the increased cortical hyperarousal experienced during N2 considering the positive relationship previously demonstrated between spindles and beta power (Fernandez and Lüthi, 2020).Sleep spindles are indicators of sleep stability (Dang-Vu et al.,  A. Brown et al. 2010).In mixed-sex studies, both time spent in N2 and spindle amplitude have been positively correlated with perceived sleep quality (Chen et al., 2023;Lok et al., 2022).This increased spindle activity may explain why we did not find significant group differences in perceived sleep quality during the current study.
Women with BSO have increased sleep apnea risk, and the role of spindle activity in affecting this risk remains unexplored (Huang et al., 2018).Work focused on midlife and older men (age 41-81 years) showed that greater spindle amplitude was associated with higher apneahypopnea index, poorer oxygen saturation < 90 % (indicative of hypoxemia), and higher arousal index (Parker et al., 2022).Increased spindle amplitude could therefore reflect enhanced thalamocortical network activity, perhaps to compensate for hypoxemia (Cruikshank et al., 2007).If there is a link between spindle amplitude and sleepdisordered breathing, the lack of increased power and amplitude in women with BSO taking ET is consistent with work showing that among women with hysterectomy and/or oophorectomy, use of any type of hormone therapy reduces sleep apnea risk (Huang et al., 2018).Only one participant in the current study's cohort was diagnosed with sleep apnea.However, sleep apnea is underdiagnosed in women, often leading to unacknowledged and untreated symptoms (Wimms et al., 2016).Future work must focus on improving sleep apnea diagnosis and awareness in women and explore the neurobiological mechanisms linked to their risk.Some sleep architecture effects observed in our study, such as increased cortical hyperarousal associated with BSO, correspond to findings from previous research focused on aging and spontaneous menopause (Campbell et al., 2011;Carrier et al., 2001).Conversely, effects such as increased spindle power and amplitude associated with BSO may be directly attributable to ovarian hormone dynamics (Driver et al., 1996).To better understand how aging and ovarian hormone loss distinctively affect sleep, it is essential for future investigations to contrast women with BSO to those in spontaneous menopause.

Spindle changes were related to verbal episodic memory performance
When all groups were combined, we found that spindle amplitude, which was greater in BSO compared to AMC, was negatively associated with verbal episodic memory performance.We have already shown that performance on this task is poorer in women with BSO (Gervais et al., 2020) and this poorer performance is related to increased sleep latency (Gervais et al., 2023).The negative link between spindle amplitude and memory in women with BSO is novel and unexpected, especially considering most mixed-sex studies show a positive association between spindle amplitude and hippocampal-dependent memory.However, spindle activity does not always correlate positively with memory.For example, a study of a mixed-sex cohort of older adults (mean age 66 years) at increased risk for dementia demonstrated spindle amplitude was associated with reduced episodic memory performance (Orlando et al., 2023).One study investigating the relationship between sleep and hippocampal-dependent memory showed that menstrual phase affected the association; performance during perimenses (low 17β-estradiol) and non-perimenses (high 17β-estradiol) was positively correlated with the number of spindles and slow oscillations, respectively (Sattari et al., 2017).These findings suggest that the relationship between spindles and memory in women is complex and likely modulated by ovarian hormones.
Sleep disorder symptomology may also affect verbal episodic memory by influencing hippocampal structure.In support of this perspective, insomnia severity in men has been linked to smaller DG and CA3 volumes (Neylan et al., 2010).Moreover, women with sleep apnea exhibit decreased posterior hippocampal DG and CA3 volumes compared to those without sleep apnea (Macey et al., 2018).Notably, work from our lab has demonstrated that women with BSO without ET have reduced volume in these same hippocampal regions (Gervais et al., 2022).Given that the hippocampus has a high density of estrogen receptors and is sensitive to fluctuating 17β-estradiol levels (Sheppard et al., 2019), it is possible that ovarian hormone loss has simultaneous direct effects on sleep disturbance and hippocampal structure, or that ovarian hormone loss affects hippocampal structure via sleep disturbance.

Strengths and limitations
Limitations to our study include its small cohort size, cross-sectional design, variations in menstrual phases for the AMC group, and diverse ET dosages, routes of administration, and timings of administration (summarized in Supplementary Material).Additionally, at-home PSG may not be as reliable as in-lab full-montage PSG.However, this is one of the few studies in younger women with BSO.Future research should confirm and expand upon our results using in-lab full-montage PSG.Our EEG configuration was optimal for assessing slow spindle activity, but future work should prioritize analyzing fast spindles with centroparietal derivations to best understand how effects may vary between slow and fast spindles.Additionally, some of our participants were taking sleep medications.These medications may influence certain sleep characteristics; benzodiazepines and some non-benzodiazepine sedative hypnotics may affect sleep spindles (Azumi and Shirakawa, 1982).However, there was not a significant group difference in the number of participants taking sleep medication and our cohort is a part of an important and under-represented clinical population, so we did not exclude participants based on the use of sleep medication.Only three participants in the cohort were taking a non-benzodiazepine as needed (BSO n = 1, BSO+ET n = 2; Table 3); therefore, it is unlikely their data heavily influenced the results.Lastly, despite all participants in the AMC group reporting regular menstrual cycling and an absence of perimenopause symptoms, one participant in this group mentioned experiencing one mild hot flash during the sleep study.Consequently, there may be a small degree of uncertainty regarding the potential inclusion of perimenopausal women in the AMC group.
Strengths of this study include that it is the first study to investigate the effects of early ovarian removal on EEG-measured sleep microarchitecture in a healthy, young group of women and to correlate it with memory performance.Findings emphasize that early BSO without ET influences various aspects of sleep macro-and microarchitecture, which could have implications for sleep disorder risk, memory, and AD progression.Another key advantage of this study was the use of a portable PSG device to evaluate sleep physiology.This method allowed for the collection of at-home sleep data over several nights, offering a more accurate representation of participants' typical sleep routines compared to a laboratory environment (Edinger et al., 1997).

Conclusions
This study supports and extends findings from our previous work investigating how early midlife BSO affects sleep and memory.We found that compared to AMC, BSO without ET was associated with insomniarelated sleep patterns, including cortical hyperarousal.Women with BSO without ET also showed increased spindle maximum amplitude, which was related to poorer hippocampal-dependent memory performance.Importantly, self-reported nocturnal hot flashes did not explain the difference in sleep architecture between groups, except in the case of delta power.Studying microarchitectural sleep changes related to BSO may provide pathways for designing sleep interventions beyond ET.However, results affirm that ET after BSO may alleviate common insomnia symptoms, including difficulty initiating and maintaining sleep (Littner et al., 2003).Given that insomnia and sleep apnea are risk factors for AD, attenuating symptoms at early midlife can have important implications for women's long-term brain health (Sadeghmousavi et al., 2020).Taken together, sleep macro-and microarchitecture changes related to BSO are evident in early midlife and relate to verbal episodic memory.This study enhances our understanding of sleep as a modifiable risk factor and how it may affect cognition preceding dementia progression.

Fig. 1 .
Fig. 1.PLS variable factor loadings for LV1 depicting how group status correlated with sleep architecture variables.Note: Latent variable 1 results for PLS analysis (p = 0.04; percent covariance accounted for 72.5 %) with 95 % confidence intervals.A) Group status factor loadings and B) Sleep architecture factor loadings.In both plots, significant items are plotted in green when the factor loading is negative and purple when it is positive, nonsignificant items are plotted in gray.The red dashed line represents the critical value which matches a p < 0.05 significance level.A variable is considered a significant contributor to the latent variable if its contribution to the latent variable exceeds what would be expected by chance, as illustrated by a factor loading surpassing the red dashed line.Abbreviations: AMC = age-matched controls, BSO+ET = bilateral salpingo-oophorectomy with current use of 17β-estradiol therapy, BSO = bilateral salpingo-oophorectomy, NREM = non-rapid eye movement sleep, TST = total sleep time, N1 = NREM stage 1, N2 = NREM stage 2, N3 = NREM stage 3, REM = rapid eye movement, log-transformed beta, delta, and spindle power (log 10 μV 2 ), spindle density = number of slow spindles per minute, slow spindle duration (seconds), slow spindle maximum amplitude (μV).

Fig. 2 .
Fig. 2. PLS variable factor loadings for LV2 depicting how group status correlated with sleep architecture variables.Note: Latent variable 2 results for PLS analysis (p = 0.04; percent covariance accounted for 27.5 %) with 95 % confidence intervals.A) Group status factor loadings and B) Sleep architecture factor loadings.In both plots, significant items are plotted in green when the factor loading is negative and purple when it is positive, nonsignificant items are plotted in gray.The red dashed line represents the critical value which matches a p < 0.05 significance level.A variable is considered a significant contributor to the latent variable if its contribution to the latent variable exceeds what would be expected by chance, as illustrated by a factor loading surpassing the red dashed line.Abbreviations: AMC = age-matched controls, BSO+ET = bilateral salpingo-oophorectomy with current use of 17β-estradiol therapy, BSO = bilateral salpingo-oophorectomy, NREM = non-rapid eye movement sleep, TST = total sleep time, N1 = NREM stage 1, N2 = NREM stage 2, N3 = NREM stage 3, REM = rapid eye movement, log-transformed beta, delta, and spindle power (log 10 μV 2 ), spindle density = number of slow spindles per minute, slow spindle duration (seconds), slow spindle maximum amplitude (μV).

Table 1
Participant demographic characteristics.

Table 2
Subjective sleep and hot flashes during polysomnography-recorded nights.

Table 3
Summary of polysomnography macro-and microarchitecture sleep measures.

Table 4
Participant sleep medication details.