Premenopausal bilateral oophorectomy leads to steeper declines in gray matter volume and alterations in perfusion and brain bioenergetics.

This is a prospective cohort study with a matched control design of healthy, cognitively normal women aged 34–55 years at entry, recruited at Weill Cornell Medicine (WMC) Departments of OBGYN Surgery and Neurology in 2021–2024 through direct referral from our surgeons or self-referral 21 , 33 , 34 . All participants provided informed consent to participate in this WMC institutional review board–approved study. All experiments were performed in accordance with relevant guidelines and regulations. At all time-points, participants underwent medical and neurological exams, laboratory analyses, neuropsychological testing, and brain imaging 21 , 33 , 34 . Pre-established exclusion criteria were medical conditions that could affect brain structure or function (e.g. stroke, neurodegenerative diseases, major psychiatric disorders, hydrocephalus, demyelinating disease such as Multiple Sclerosis, intracranial mass, and infarcts on MRI), use of psychoactive medications, and MRI contraindications. None of the participants had a history of cancer or cancer-related treatments such as chemotherapy or adjuvant endocrine therapy. Participants had baseline Montreal Cognitive Assessment (MoCA) ≥ 26 and normal cognitive performance for age and education. Twelve women scheduled to undergo bilateral oophorectomy prior to menopause, with or without hysterectomy, for benign or prophylactic indications (endometriosis, uterine fibroids, adnexal mass, or prophylactic surgery due to family history of ovarian cancer) agreed to participate. None of the participants were MHT users. Participants underwent the baseline MRI session a median of 0.04, IQR 0.03–0.67 months before the procedure. Participants completed follow-up sessions at 1.8 ± 1.3 years post-surgery. Five had an additional follow-up at 2.7 ± 1.6 years, and one had a third follow-up 6 months later. We paired each PO participant with two controls (2:1 ratio) from our active database to enhance statistical power while maintaining feasibility. Matching was conducted through direct selection based on baseline age (± 2 years), menopausal stage (~ 1:1 pre- or perimenopausal), education (± 1 year), race (1:1 minority representation), time to follow-up (± 6 months), and MHT non-user status. The final sample consisted of 12 PO patients and 24 controls. In one PO participant, the pre-surgical MRI contained movement artifacts, and a prior MRI obtained 28 months before surgery was used instead. Analyses were repeated after excluding this participant, which left results largely unchanged. At the time of study design, no prior longitudinal neuroimaging data in PO populations were available to inform effect size estimates for power calculations. As such, we conducted a post-hoc approximation of statistical power using G*Power 3.1 based on the available sample size ( n  = 36; 12 PO, 24 controls). These estimates should be interpreted as upper bounds under idealized conditions. Under these simplified assumptions, the study was sufficiently powered to detect large effects (Cohen’s f 2 ≈0.35) and modestly powered for medium-sized effects ( f 2 ≈0.15) for cross-sectional group differences and group-by-time interaction effects in linear models. For non-linear (quadratic) effects, estimated power was adequate for large effects but limited for medium or small effects. Accordingly, we restricted interpretation to effects of large magnitude and strong model support (e.g., lower AIC values), and we conservatively frame and interpret our results as hypothesis-generating rather than confirmatory. Participants underwent a cognitive testing battery including the Rey Auditory Verbal Learning Test (RAVLT) for encoding and retention, the Logical Memory subtest of the Wechsler Memory Scale for narrative episodic memory, the Trail Making Test part B (TMT-B) for executive function, F-A-S for verbal fluency, and the Boston naming test (BNT) for language function 21 , 33 , 34 . Determination of menopausal status was based on the Stages of Reproductive Aging Workshop (STRAW-10) criteria 47 with hormone assessments as supportive criteria 26 . Participants were classified as premenopausal (regular cycler), perimenopausal (irregular cyclers with interval of amenorrhea ≥ 60 days or ≥ 2 skipped cycles) and postmenopausal (no cycle for ≥ 12 months) 26 . Only women with a menstrual cycle at baseline were included in this study. Information on hormone therapy usage was obtained through the interview with the examining physician. Before each brain imaging session, participants underwent a blood draw by venipuncture after an overnight fast. Samples were shipped overnight to CLIA-certified Boston Heart Diagnostics [Framingham, MA] and analyzed on a Roche Cobas e801 analytical unit for immunoassay tests using Electrochemiluminescence technology (ECL) [Roche Diagnostics; Basel, Switzerland]. FSH and LH were assessed through electrochemiluminescence sandwich immunoassay (ELISA) [measuring range 0.3–200 mIU/mL], and estradiol (E2) through competitive immunoassay [measuring range 5–3000 pg/mL]. At each time point, participants received three MR sequences on a 3.0 Tesla MR750 Discovery scanner (General Electric, Waukesha, WI) equipped with a 32-channel head coil, using the same protocol and equipment, under published protocols 21 , 33 , 34 . These included volumetric T 1 -weighted BRAVO MRI [1 × 1 × 1 mm resolution, 8.2 ms repetition time (TR), 3.2 ms echo time (TE), 12° flip angle, 25.6 cm field of view (FOV), 256 × 256 matrix with ARC acceleration], ASL [pseudo-continuous technique with 4851 ms TR, 10.6 ms TE, 4 averages, 24 cm FOV, 2.0 × 2.0 × 3.8 mm resolution] 48 , and 31 P-MRS acquired on the same scanner as the MRI, typically on the same day, using a dual tuned 32-channel 31 P/ 1 H quadrature head coil (Clinical MR Solutions, Brookfield, WI) [2048 points, 5000 Hz sweep width, 2000 ms TR, 2 averages, 55° flip angle at 51.3 MHz, 24 cm FOV] 34 , 35 , 49 . For 31 P-MRS, a 3 Plane Localizer image with 20 images in each orthogonal direction was acquired. Shimming was performed using a 1 H single voxel technique placed over the entire brain. A high-resolution, 8-slice, 5 mm slice thickness, sagittal T 2 -Fluid Attenuated Inversion Recovery sequence (FLAIR) was acquired at the same location as the 31 P-MRS CSI slices. MRS data were processed using XSOS using Hamming and Fermi k-space filters, 20 Hz exponential filtering and zero-filling in time, x and y-domains prior to 3D Fast Fourier Transformation 34 , 35 , 49 . Peak area integration was performed for phosphocreatine (PCr) and total ATP (α-ATP, β-ATP and γ-ATP moieties) 34 , 35 , 49 by an experienced analyst (JPD). Two participants did not complete MRS, and one did not complete ASL scanning due to technical issues. We used the Normalized Mutual Information routine of Statistical Parametric Mapping (SPM12) 50 implemented in Matlab R2023b (MathWorks; Natick, MA) to align each participant’s T 1 BRAVO to the corresponding ASL and reference T 2 -FLAIR scan, and then align the MRS maps with the skull stripped MRI using the Normalized Mutual Information routine 50 . Scans were quantified using FreeSurfer 7.2 28 and Desikan-Killiany Atlas-based regions of interest (ROI) 29 applied to the aligned MRI. We focused on bilateral ROIs with known neuroendocrine aging vulnerability 10 : hippocampus, parahippocampal gyrus, hypothalamus, and global white matter for volumetric analyses; medial temporal lobe (average of hippocampus, amygdala, entorhinal and parahippocampal gyrus), middle and superior frontal gyrus for CBF and PCr/ATP measures. Total intracranial volume (TIV) was obtained for normalization purposes. Analyses were performed using R v.4.5.0 and SPM12, with supplemental models run in SPSS v.29. Python 3.11 was used for data visualization. For clinical measures, categorical variables were compared using Fisher’s exact test, and continuous variables were compared using Mann-Whitney U or Wilcoxon tests, at p  < 0.05. Mixed-effects models were employed to investigate changes in ROI biomarkers with a two-level exposure (PO vs. controls). Time was modeled continuously as months since surgery or since the matched index date for controls. For controls, we defined an “index date” equivalent to the surgical date for their matched PO participant. Follow-up months were then calculated relative to this index to allow alignment of time variables across groups. All models included random intercepts to account for within-subject correlations across repeated measures. The analytic approach was implemented in two sequential steps: (i) Model 1 : We fit linear mixed-effects models that included fixed effects for time, group, and time × group interactions, using restricted maximum likelihood with a random intercept for participant to account for within-subject correlation; (ii) Model 2 : Quadratic time terms (time 2 ) were added to the models to assess potential non-linear trajectories in outcomes, testing for fixed effects for time (linear and quadratic), group, and their interactions. For models yielding significant quadratic estimates, we compared linear and quadratic mixed-effects models to determine the best fit using Akaike Information Criterion (AIC). Degrees of freedom and p-values for fixed effects were computed using the Satterthwaite approximation as implemented in the lmerTest package. To aid interpretability, we additionally report approximate standardized effect sizes (Cohen’s d) for interaction terms derived from regression coefficients and their standard errors (β/SE). These values are intended as descriptive indices of effect magnitude rather than formal variance-explained metrics, as commonly done in small-sample mixed-model studies where partial R 2 estimates are unstable or unavailable 51 . Effect sizes of 0.2, 0.5, and 0.8 were interpreted as small, medium, and large, respectively 52 . Although the groups did not differ significantly in mean age, analyses were adjusted for baseline age as a covariate given the wide age range (35–55 years), to account for possible age-related variance in biomarker outcomes. Modality-specific confounders modeled as covariates were total intracranial volume (TIV) for volumetric MRI, and global mean CBF for perfusion measures. 31P-MRS ATP measures are normalized to PCr and expressed as ratios. All models were estimated using restricted maximum likelihood (REML), and significance was assessed using Type III Wald F-tests with an alpha threshold of 0.05, uncorrected. Estimated marginal means and predicted trajectories with 95% confidence intervals (CI) were extracted from significant models for interpretability. Interaction terms were the primary focus of hypothesis testing. Effects of age at surgery . To examine whether younger age at surgery was associated with steeper or more nonlinear biomarker trajectories within the PO group, follow-up models were conducted on PO participants and included interaction terms for both linear and quadratic time by age at surgery. Associations with plasma hormone levels. Target plasma hormone measures (estradiol, FSH, and LH) were log-transformed to normalize their distributions. The mixed-effects modeling approach described above was applied to assess effects of group × time on hormone trajectories, at p  < 0.05. Secondly, log-transformed plasma hormone concentrations were modeled as time-varying predictors of ROI measures over time, and group status and its interaction terms with hormone predictors (group × hormone) were included to test whether the relationships between hormone levels and brain biomarkers varied by group. Testing of associations between plasma hormones and brain biomarkers was restricted to measures showing significant interaction effects. Associations with cognitive performance. The mixed-effects modeling approach described above was applied to assess group × time effects on cognitive trajectories, at p  < 0.05. Testing of associations between brain biomarkers and cognition was restricted to measures showing significant interaction effects.

Participants’ characteristics are presented in Table  1 . The sample included 12 women who underwent PO and 24 matched controls (see Online Methods for inclusion/exclusion criteria). The average follow-up period was 2.7 years (SD = 1.6) overall. Among PO participants, 5 (41.7%) had coexisting diagnoses of endometriosis and uterine fibroids, 5 (41.7%) had fibroids only, and 2 (16.7%) underwent prophylactic bilateral oophorectomy due to a family history of ovarian cancer (Table  1 ). Table 1 Patient characteristics. N Premenopausal oophorectomy (PO) Controls P value a 12 24 Age at baseline, years, range 48(6), 35–55 49(5), 34–55 0.439 Education, years 18(17,19) 17(16,18) 0.110 Race, % white 67 67 1.000 Menopause status at the time of the baseline exam, % premenopause/perimenopause b 33 / 67 38/62 0.727 APOE-4 status, % positive 17 38 0.268 Time to follow-up, years, range 2.3(1.9), 1.0-4.9 2.8(1.3), 1.5–6.2 0.410 Time between surgery and imaging, months 0.04(0.03,0.67) n.a. Surgical indications, %: n.a.  Family history of ovarian cancer 16.7  Fibroids, abnormal uterine bleeding 16.7  Fibroids 25.0  Fibroids, endometriosis 41.7 Cognitive test scores, unitless c :  MoCA 28(27,29) 28(27,30) 0.559  Logical memory, immediate 14.33(4.10) 15.09(4.09) 0.505  Logical memory, delayed 13.58(4.72) 13.29(5.32) 0.932  RAVLT, total 27.92(4.78) 27.09(9.29) 0.921  RAVLT, delayed recall 9.17(2.29) 11.21(2.54) 0.393  RAVLT, recognition 13.42(1.44) 13.45(1.74) 0.777  F-A-S (fluency test) 50.17(7.61) 42.09(14.83) 0.168  Animal naming 22.50(5.27) 23.09(7.62) 0.606  Boston naming 14.08(1.51) 14.36(1.22) 0.639  Trail making test (TMT)-B 50.91(21.49) 42.78(17.56) 0.403 N, %; continuous measures are presented as mean (SD) for normally distributed variables and as median (IQR) for non-normally distributed variables. Montreal Cognitive Assessment, MoCA; Ray Auditory Verbal Learning test, RAVLT. a Mann-Whitney U or Wilcoxon rank sum test; Fisher’s exact test. b PO group: menopause status at the time of oophorectomy. c Mean (SEM) from linear regressions adjusted by age and education. Patient characteristics. N, %; continuous measures are presented as mean (SD) for normally distributed variables and as median (IQR) for non-normally distributed variables. Montreal Cognitive Assessment, MoCA; Ray Auditory Verbal Learning test, RAVLT. a Mann-Whitney U or Wilcoxon rank sum test; Fisher’s exact test. b PO group: menopause status at the time of oophorectomy. c Mean (SEM) from linear regressions adjusted by age and education. As described in the Online Methods , primary analyses focused on preselected bilateral ROIs with known vulnerability to neuroendocrine aging 10 extracted using FreeSurfer 7.2 28 and Desikan-Killiany Atlas-based regions of interest (ROI) 29 . Target ROIs included hippocampus, parahippocampal gyrus and hypothalamus for GMV; global white matter for WMV; medial temporal and frontal cortices (middle and superior frontal gyri) for CBF and PCr/ATP. Modality-specific confounders modeled as covariates were total intracranial volume (TIV) for volumetric measures and global mean CBF for perfusion analyses. 31 P-MRS ATP measures were normalized to PCr and expressed as ratios. All primary models adjusted for baseline age at enrollment as a covariate. Gray matter volumes : Results are presented in Table  2 . There were no significant baseline group differences in any region examined. Adjusting by age and TIV, group × time interactions were observed for hippocampal ( p  < 0.001) and parahippocampal volumes ( p  = 0.008). In both regions, the PO group exhibited steeper volume declines compared to controls (Fig.  1 ), corresponding to medium to large estimated effect sizes (Cohen’s d : hippocampus = − 0.62, parahippocampus = − 0.48). Addition of quadratic time terms did not improved model fit or reveal evidence of nonlinear change in these regions ( p >  0.22). No linear or nonlinear effects were found for hypothalamic volume ( p  > 0.30). Table 2 Longitudinal effects of premenopausal oophorectomy (PO) status on regional brain volume. Region Parameter Model 1 Model 2 Estimate Std. error P -value Estimate Std. error P -value Hippocampus Group (PO) 188.81 92.78 0.050 120.52 96.97 0.221 Time 3.816 0.991 < 0.001 3.645 1.175 0.003 Group × time − 4.587 1.228 < 0.001 − 4.189 1.712 0.019 Time 2 0.011 0.041 0.783 Group × time 2 − 0.026 0.078 0.740 Parahippocampal gyrus Group (PO) 69.31 110.09 0.534 21.51 112.51 0.850 Time 1.513 0.8666 0.086 1.668 0.906 0.070 Group × time − 1.662 0.584 0.008 − 2.339 0.795 0.006 Time 2 − 0.013 0.021 0.546 Group × time 2 0.047 0.038 0.224 Hypothalamus Group (PO) 33.28 16.31 0.049 36.45 16.96 0.038 Time 0.270 0.1686 0.114 0.222 0.195 0.259 Group × time − 0.132 0.2001 0.515 0.064 0.275 0.817 Time 2 0.003 0.007 0.648 Group × time 2 − 0.013 0.013 0.303 Global white matter Group (PO) 8797.29 11674.49 0.462 14864.35 10869.59 0.182 Time − 71.13 109.39 0.519 − 244.08 111.73 0.032 Group × time − 40.07 110.58 0.721 196.07 141.86 0.177 Time 2 12.84 3.51 < 0.001 Group × time 2 − 16.32 6.53 0.018 Results from linear (Model 1) and non-linear (Model 2) mixed effects models adjusting by age and total intracranial volume. Significant interaction effects are in bold. Longitudinal effects of premenopausal oophorectomy (PO) status on regional brain volume. Results from linear (Model 1) and non-linear (Model 2) mixed effects models adjusting by age and total intracranial volume. Significant interaction effects are in bold. Fig. 1 Predicted trajectories of regional brain volume changes by group. Group-level predicted trajectories for standardized change from baseline are shown for hippocampal, parahippocampal, hypothalamic, and global white matter volume over time since surgery or time since index date for controls (months). Standardized predictions follow the same model structure used in primary analyses, with outcome variables normalized to baseline values, adjusted for age and total intracranial volume. Models included linear and quadratic terms for time and group-by-time interactions where appropriate. Predicted trajectories with 95% confidence intervals derived from full model variance (fixed effects, random effects, and residuals) were visualized over a 60-month period using Python 3.11 (matplotlib). For visualization, constant confidence bands were applied across time, using group- and region-specific half-widths. Premenopausal oophorectomy (PO) group shown in red; controls shown in gray. Predicted trajectories of regional brain volume changes by group. Group-level predicted trajectories for standardized change from baseline are shown for hippocampal, parahippocampal, hypothalamic, and global white matter volume over time since surgery or time since index date for controls (months). Standardized predictions follow the same model structure used in primary analyses, with outcome variables normalized to baseline values, adjusted for age and total intracranial volume. Models included linear and quadratic terms for time and group-by-time interactions where appropriate. Predicted trajectories with 95% confidence intervals derived from full model variance (fixed effects, random effects, and residuals) were visualized over a 60-month period using Python 3.11 (matplotlib). For visualization, constant confidence bands were applied across time, using group- and region-specific half-widths. Premenopausal oophorectomy (PO) group shown in red; controls shown in gray. Global white matter volume : WMV showed no baseline group differences ( p  = 0.477). Interaction effects indicated divergence in trajectories following surgery, with a decline in the PO group (time: β = − 244.1, p  = 0.032; time 2 : β = 12.84, p  < 0.001) and a group × time 2 interaction (β = − 16.32, p  = 0.018), corresponding to a large estimated effect size ( d  = − 2.50) (Table  2 ; Fig.  1 ). The quadratic model showed better fit than the linear model (Akaike Information Criterion, AIC 1628.44 vs. 1646.82; ΔAIC = 18.38), supporting accelerated WMV loss in the PO group compared to controls. Cerebral blood flow (CBF) : Results are presented in Table  3 . No baseline group differences were observed in any region examined. Longitudinally, the PO group exhibited increasing superior frontal CBF relative to controls (group × time: β  = 0.097, p  = 0.039), with a group × time 2 interaction ( β  = − 0.0042, p  = 0.036), corresponding to a large estimated effect size ( d  = 2.5). The quadratic model provided a better fit than the linear model (AIC 421.84 vs. 442.70; ΔAIC = 20.86), suggesting a nonlinear trajectory marked by early CBF increase followed by a plateau or decline in the PO group compared to controls (Fig.  2 ). A similar effect was observed in the middle frontal region (group × time 2 : β  = − 0.048, p  = 0.049), corresponding to a large estimated effect size ( d  = 2.0) (Fig.  2 ). However, model fit did not favor a quadratic over the linear specification (434.93 vs. 456.70; ΔAIC = 21.77), and this finding was therefore not considered statistically robust. No significant effects were observed for medial temporal CBF in either model ( p  > 0.52). ATP production (PCr/ATP) : Results are presented in Table  3 . No baseline group differences were observed in any region examined. Time was a significant predictor of PCr/ATP increases in superior and middle frontal regions ( p   ≤  0.045), indicating a modest increase in metabolic demand over time, with no differential effect of PO status. We note a trend-level group × time 2 interaction in the medial temporal region ( p  = 0.077), suggesting a borderline steeper early increase followed by decline in the PO group compared to controls. Table 3 Longitudinal effects of premenopausal oophorectomy (PO) status on regional CBF and PCr/ATP. Biomarker Region Parameter Model 1 Model 2 Estimate Std. error P -value Estimate Std. error P -value CBF Medial temporal Group (PO) − 17.09 15.84 0.288 − 16.73 15.86 0.299 Time 0.215 0.118 0.074 0.156 0.127 0.224 Group × time − 0.038 0.083 0.648 0.013 0.115 0.911 Time 2 0.003 0.003 0.224 Group × time 2 − 0.003 0.005 0.521 Middle frontal Group (PO) 2.14 1.60 0.189 4.50 1.51 0.004 Time − 0.007 0.022 0.726 − 0.014 0.032 0.654 Group × time − 0.036 0.035 0.317 0.092 0.056 0.107 Time 2 0.001 0.001 0.640 Group × time 2 − 0.005 0.002 0.049 Superior frontal Group (PO) 0.721 1.295 0.581 3.02 1.45 0.043 Time − 0.017 0.021 0.406 − 0.017 0.026 0.524 Group × time 0.027 0.035 0.443 0.097 0.045 0.039 Time 2 − 0.000 0.001 0.950 Group × time 2 − 0.004 0.002 0.036 PCr/ATP Medial temporal Group (PO) 0.069 0.048 0.155 0.135 0.05 0.014 Time 0.002 0.001 0.022 0.001 0.001 0.579 Group × time 0.000 0.002 0.839 0.003 0.002 0.152 Time 2 0.000 0.005 0.090 Group × time 2 − 0.001 0.000 0.077 Middle frontal Group (PO) 0.104 0.087 0.239 0.116 0.116 0.248 Time 0.004 0.002 0.025 0.006 0.006 0.045 Group × time 0.000 0.003 0.859 0.001 0.005 0.912 Time 2 0.000 0.000 0.522 Group × time 2 − 0.000 0.000 0.975 Superior frontal Group (PO) 0.118 0.111 0.293 0.103 0.124 0.408 Time 0.004 0.002 0.058 0.006 0.003 0.038 Group × time 0.000 0.003 0.848 − 0.000 0.005 0.909 Time 2 − 0.000 0.000 0.263 Group × time 2 0.000 0.000 0.745 Results from linear (Model 1) and non-linear (Model 2) mixed effects models adjusting by age and modality-specific confounders. Significant interaction effects are in bold. CBF, cerebral blood flow; PCr/ATP, phosphocreatine to adenosine triphosphate ratios; PO, premenopausal oophorectomy. Longitudinal effects of premenopausal oophorectomy (PO) status on regional CBF and PCr/ATP. Results from linear (Model 1) and non-linear (Model 2) mixed effects models adjusting by age and modality-specific confounders. Significant interaction effects are in bold. CBF, cerebral blood flow; PCr/ATP, phosphocreatine to adenosine triphosphate ratios; PO, premenopausal oophorectomy. Fig. 2 Predicted trajectories of regional cerebral blood flow changes by group. Group-level predicted trajectories for standardized change from baseline are shown for superior frontal, middle frontal, and medial temporal cerebral blood flow (CBF) over time since surgery or time since index date for controls (months). Standardized predictions follow the same model structure used in primary analyses, with outcome variables normalized to baseline values, adjusted for age and global CBF. Models included linear and quadratic terms for time and group-by-time interactions where appropriate. Predicted trajectories with 95% confidence intervals derived from full model variance (fixed effects, random effects, and residuals) were visualized over a 60-month period using Python 3.11 (matplotlib). For visualization, constant confidence bands were applied across time, using group- and region-specific half-widths. Premenopausal oophorectomy (PO) group shown in red; controls shown in yellow. Predicted trajectories of regional cerebral blood flow changes by group. Group-level predicted trajectories for standardized change from baseline are shown for superior frontal, middle frontal, and medial temporal cerebral blood flow (CBF) over time since surgery or time since index date for controls (months). Standardized predictions follow the same model structure used in primary analyses, with outcome variables normalized to baseline values, adjusted for age and global CBF. Models included linear and quadratic terms for time and group-by-time interactions where appropriate. Predicted trajectories with 95% confidence intervals derived from full model variance (fixed effects, random effects, and residuals) were visualized over a 60-month period using Python 3.11 (matplotlib). For visualization, constant confidence bands were applied across time, using group- and region-specific half-widths. Premenopausal oophorectomy (PO) group shown in red; controls shown in yellow.

These data provide first-time evidence that PO alters brain biomarker trajectories in some HPG axis regions compared to normal aging, providing a framework for increased long-term neurological vulnerability. We frame our results as hypothesis-generating and caution that these observations are based on a small, carefully screened sample and emphasize the need for replication in larger, more diverse cohorts.

In this prospective, longitudinal neuroimaging study of midlife women undergoing PO for non-cancerous conditions, over an average 2.9 ± 1.7 year period, PO led to steeper declines in hippocampal and parahippocampal volume, steeper and accelerating declines in white matter volume, and accelerated increases in superior frontal CBF compared to controls. In exploratory analyses, LH and FSH increased at significantly higher rates in the PO group than in controls, which is an expected physiological finding. Higher LH/FSH were associated with lower hypothalamic and parahippocampal volume and higher medial temporal CBF relative to controls. Limited previous work has specifically examined the impact of PO on the brain. At post-mortem, PO before age 45 was associated with a higher burden of neuritic plaques, a hallmark of Alzheimer’s disease (AD) 22 . Among in vivo studies, two analyses of the Mayo Clinic Study of Aging (MCSA) cohort reported smaller amygdala volumes, thinner parahippocampal-entorhinal cortices, and lower entorhinal white matter integrity in the PO group compared to controls 16 , 19 . However, brain imaging occurred ~ 19 years post-surgery, and nearly all patients (81–96%) had used MHT 16 , 19 . A cross-sectional study of BRCA1/2 mutation carriers who received MRI ~ 5 years post-PO found that those who did not receive estrogen therapy (ET) had smaller bilateral dentate gyrus/CA2/CA3 volumes compared to premenopausal controls, whereas those treated with ET did not 17 . Additional analyses of this cohort showed greater perirhinal BA 36 volume particularly in the PO group without ET 18 . Generally, the reliance on retrospective designs, the inclusion of women on MHT, and the time intervals between PO and imaging preclude determination of whether these effects are directly attributable to PO. Our prospective study expands prior literature by providing evidence of neurophysiological changes following PO in women who had not initiated MHT, and establishing a timeline for risk emergence within three years post-surgery. These results have several implications. While no group differences were observed at baseline, the PO group exhibited progressive hippocampal and parahippocampal volume loss not observed in controls. These regions are rich in sex steroid hormone receptors and are tightly coupled to cognitive function and emotion regulation 8 – 10 . These findings are consistent with prior evidence of smaller hippocampal sub-volumes within 5 years post-BSO 17 , and thinner entorhinal-parahippocampal cortices in older PO women 16 . Further, PO led to accelerated WMV loss, which in exploratory analyses was more pronounced in association with earlier age at surgery. Another novel result is the evidence for nonlinear effects on CBF in superior frontal regions, suggesting a transient increase followed by a decline or plateau in the PO group compared to controls. These findings may reflect a biphasic response, with transient hyper-perfusion followed by a breakdown or stabilization of vascular compensation. In contrast, CBF in the medial temporal lobe exhibited no differential group effects using ROI, though differential increases were noted in the left hemisphere of the PO group using VBA, which warrant further investigation. While PCr/ATP changes did not significantly vary by group, in exploratory analyses, a younger age at PO was associated with a steeper early increase followed by sharper declines in mitochondrial energy production in medial temporal and middle frontal regions. Overall, the co-occurrence of reduced volume in limbic regions and white matter, along with increased frontal CBF, suggests a concerted response to PO-induced neuronal stress. Limbic volume loss can disrupt afferent projections to frontal cortex, resulting in compensatory hyper-perfusion and increased functional recruitment of those regions 30 . Elevated PCr/ATP in limbic and frontal areas in younger PO patients (a sign of inefficient energy utilization) suggest that earlier hormonal disruption is metabolically costly. Together, these findings may represent increased reliance on frontal circuits to support limbic function, albeit under constrained energy demands, as the brain attempts to maintain cognition under altered neuroendocrine conditions. This aligns with preclinical evidence that menopause-related hormonal changes trigger cerebral glucose hypometabolism 31 and increased ketone body utilization as an alternative ATP source 32 , in turn leading to compromised mitochondrial efficiency, white matter catabolism, and cellular apoptosis 31 , 32 . Findings reported herein are also consistent with significant decline in WMV, and CBF and ATP alterations in women undergoing spontaneous menopause 21 , 33 – 35 . While the precise biological mechanisms underlying nonlinear trajectories in brain biomarkers following oophorectomy remain to be fully elucidated, evidence from both animal models and human studies suggests that the associated abrupt hormonal changes may accelerate aging-related processes across multiple organ systems 10 – 12 . In preclinical studies, surgical menopause has been associated with accelerated decline in cardiovascular, skeletal, and immune function, as well as changes in neuroinflammation and myelin repair pathways 36 , 37 . Mechanistic analyses have shown associations between bilateral oophorectomy and accelerated aging as measured by epigenetic biomarkers, suggesting that the premature loss of ovarian function may lead to an increase in the extent of DNA methylation, a biological marker of accelerated aging 11 . Consistent with these observations, clinical studies show that women who undergo early or surgical menopause face elevated risks for cardiovascular disease, osteoporosis, and cognitive decline 13 – 15 . These findings support the broader hypothesis that loss of ovarian hormones may disrupt homeostatic regulation and promote non-linear aging trajectories in certain tissues—including the brain. Both estradiol depletion and gonadotropin increases have demonstrated neurotoxic effects 10 , 38 which may account for neurological disruption following PO. In this study, LH and FSH increased more rapidly in the PO group compared to controls, and these increases were associated with greater declines in hypothalamic and parahippocampal volume, respectively, and with increased medial temporal CBF relative to controls. This is consistent with literature implicating the abrupt rise in gonadotropins as a driver of neuropathological aging post-PO 38 – 40 . PO status was associated with overall reduced estradiol levels, consistent with ovarian removal, and younger age at PO was associated with steeper estradiol declines. The rate of change did not significantly differ between groups, which may be due to a combination of consistently low estradiol levels in the PO group and variability in the timing of blood draws among controls, which were not standardized by menstrual cycle phase. As this may introduce variability in hormone levels unrelated to group or time effects, plasma hormone analyses are presented as sensitivity analyses, rather than primary outcomes. Despite this limitation, the observed associations are consistent with biological expectations—such as increased gonadotropins following PO and their associations with specific brain regions—supporting their plausibility and interpretability within the study’s exploratory framework. To note, the observed hormone-brain associations are correlational and cannot establish causality or confirm a temporal sequence. Prior literature supports the hypothesis that surgical menopause leads to hormonal shifts, which in turn influence neurobiological aging 10 – 12 , suggesting a role for these hormones as upstream regulators rather than downstream effects. However, bidirectional or recursive dynamics are plausible given known feedback loops between central and peripheral systems. Future studies incorporating tighter temporal resolution, multimodal biomarker integration, and potentially interventional designs are needed to disentangle these complex relationships. This study has the following strengths. First, we conducted pre- vs. post-surgery assessments in participants undergoing PO for benign or prophylactic indications, with clinical, cognitive, laboratory exams, and high-resolution neuroimaging. We focused on cancer-free individuals to avoid possible confounding effects of cancer and cancer treatments. Second, patients were retained in the study for an average of two years post-surgery, while undergoing repeated follow-ups. Third, each patient was matched with two control participants without a history of oophorectomy/hysterectomy, who remained pre- or peri-menopausal throughout the study period, all MHT non-users. This design enabled us to examine putative direct links between PO and brain biomarker changes. Consistent with guidelines recommending ovarian preservation in premenopausal women undergoing oophorectomy for non-cancerous indications 41 , our PO cohort was enriched for women closer to the typical age of menopause, with a mean age of 48 years, and 1/3 under age 46. Notably, one participant remained perimenopausal at age 55. Control matching accounted for menopausal status, and the age range allowed us to explore associations with age at PO. It remains to be determined whether brain changes would be more pronounced following PO in younger women. While months since surgery was used as the time variable for the PO group, the control group lacked a comparable event to anchor longitudinal change. Therefore, the date of the baseline exam was used as the index date for controls, with groups matched on baseline age and menopausal status (pre- or peri-menopausal). While this approach allows alignment of time variables across groups, it may introduce variability related to unmeasured differences in the timing of underlying neuroendocrine transitions. A main limitation of this study is the small sample size, reflecting the inherent challenges of recruiting for prospective pre-to-post-surgical neuroimaging in this population. To maximize statistical efficiency, we employed a carefully matched cohort and used mixed-effects models, which are well-suited to handle irregular follow-up intervals and allow inclusion of all available observations. Nonetheless, the limited sample size and relatively short follow-up duration constrain the generalizability of our findings, warranting further validation in larger-scale, longer-term studies. We have therefore framed all results as exploratory and hypothesis-generating, rather than confirmatory of hypothesis-testing, intended to provide preliminary effect size estimates and inform the design of future larger-scale, longer-term studies. The study was sufficiently powered to detect moderate-to-large linear effects and large quadratic effects—changes likely to be of clinical or biological relevance. As such, we only reported effects that fall within the range of detectable effect sizes, and we restricted interpretation of nonlinear effects to those showing large magnitude and strong model support (e.g., lower AIC). Overall, findings related to acceleration or curvature in biomarker trajectories should be viewed as preliminary and hypothesis-generating, pending replication in larger samples. Given the exploratory nature of this study and the limited sample size, we did not apply a single global correction across all outcomes. Instead, each imaging modality (volumetric MRI, CBF, and 31 P-MRS) was analyzed independently based on pre-specified, biologically informed hypotheses, with a limited number of targeted outcomes (1–3 regions) per modality. Because these modalities are technically and physiologically distinct, they were treated as separate analytic families, thereby reducing the risk of inflated Type I error associated with large-scale multiple testing. For discussion purposes, we applied a relaxed Benjamini–Hochberg false discovery rate (FDR) correction 42 within each modality, except for WMV, which comprised a single outcome. Under this approach, significant group × time interactions for hippocampal and parahippocampal volumes remained significant (q < 0.05), whereas CBF interaction effects were nominal and did not survive correction. These results reinforce our interpretation of volumetric findings as more robust, while supporting a cautious, hypothesis-generating interpretation of perfusion-related nonlinear effects. Another limitation of the current study is the limited generalizability of our findings to the broader population of women undergoing PO for varying clinical indications. The majority of PO participants in our cohort underwent surgery for endometriosis and/or fibroids, with two participants undergoing prophylactic oophorectomy. While this reflects the clinical population most accessible for prospective neuroimaging research, it restricts our ability to disentangle indication-specific effects. For instance, the higher prevalence of endometriosis in the PO group may have influenced baseline hormonal profiles and outcomes. Future studies in larger and more heterogeneous cohorts are warranted to conduct age stratification and subgroup analyses by surgical indication, test for indication-specific hormonal variations, and to assess the generalizability of these findings across surgical populations. Although longitudinal group differences in hypothalamic volume were not significant, the PO group showed a greater rate of decline than controls, and this decline was associated with increasing LH levels. Lack of group effects may reflect segmentation variability in this small region, combined with limited sample sizes. However, detection of expected associations with circulating LH levels provides support for the biological validity of the measurements despite the absence of group effects. In this study, we used FreeSurfer v.7.2, which incorporates a validated probabilistic atlas for automated subcortical segmentation, including the hypothalamus. While the fully automated nature of this approach precluded intra- or inter-rater reliability assessments, we recognize this as a limitation. Given the hypothalamus’s central role in neuroendocrinology, studies with larger samples are needed to evaluate whether differences become significant over time. Similarly, no group differences were observed in cognitive performance. This may reflect the young age and high educational attainment of our participants and the fact that all were cognitively normal at baseline, with MoCA scores ≥ 26, which could confer cognitive reserve and reduce sensitivity to early changes. While our neuropsychological battery included gold-standard cognitive tests, with a focus on domains previously shown to be estrogen-sensitive, such as verbal memory, the well-documented female advantage in verbal memory 43 may reduce the utility of these measures for detecting subtle decline in women. Ceiling effects may also have contributed limited our ability to detect differential effects. Future studies are needed to determine whether more sensitive or domain-specific measures might better capture potential differences. Alternatively, the absence of cognitive deficits or alterations despite measurable post-PO changes in brain structure and perfusion suggests the presence of compensatory mechanisms that preserve function in the face of neurobiological change—at least over a ~ 2-year follow-up period. Continued follow-up is warranted to determine whether the observed brain alterations may translate into clinically meaningful cognitive outcomes in the future. These preliminary findings highlight several promising avenues for future research. A natural extension of this work will be to examine the effects of MHT on PO’s induced brain changes. In absence of contraindications, MHT is recommended for women undergoing PO, in part to support cognition and mood 44 , and prior studies suggest that ET initiated near the time of oophorectomy may support memory and reduce AD risk 45 , 46 . Future trials are warranted to test whether MHT can mitigate neurobiological aging in women undergoing PO, especially when initiated early. Future research should aim to characterize the multidimensional effects of surgical menopause through complementary biomarker approaches, spanning AD disease biomarkers (amyloid-beta, tau), neuroinflammatory and neurovascular indices, multi-omics profiling, and task-based fMRI, to delineate the functional implications of observed brain changes. Longer-term follow-up will be essential to determine whether early brain changes observed here ultimately translate into cognitive decline.

Baseline LH and FSH levels did not differ between groups. However, group × time interactions indicated more rapid increases in gonadotropins in the PO group compared to controls (LH β = 0.547, SE = 0.143, p  < 0.001, d = 0.64; FSH β = 0.417, SE = 0.184, p  = 0.029, d = 0.38) (Supplementary Fig.  S1 , Supplementary Table  S1 ). Baseline estradiol levels were lower in the PO group compared to controls (β = − 1.257, p  = 0.009), with no significant group × time effects (Supplementary Table  S1 ). Plasma hormones showing significant differential effects were tested for associations with brain biomarkers. Group × LH interactions were observed for hypothalamic volume (β = − 13.60, p  = 0.043) and medial temporal CBF (β = 6.32, p  = 0.029), while group × FSH interactions emerged for parahippocampal volume (β = − 70.02, p  = 0.005) and medial temporal CBF (β = 6.90, p  = 0.032), with marginal effects observed for hypothalamic volume (β = − 14.12, p  = 0.058) (Supplementary Table  S2 ). No associations were found with PCr/ATP measures. Given prior associations between younger age at menopause and higher neurological risk 13 – 15 , age at surgery—rather than chronological age—was examined as a predictor of biomarker outcomes among PO participants (Table 4 ). Younger age at PO was associated with steeper increases and sharper declines in WMV over time (time × age: β = − 106.2, p = 0.024; time 2 × age: β = 2.51, p = 0.049, d = 0.39). Younger age at PO was also associated with greater curvature in PCr/ATP in medial temporal (time × age: β = − 0.002, p = 0.013; time 2 × age: β = 4.9e–5, p = 0.017, d = 0.17) and middle frontal regions (time × age: β = − 0.003, p = 0.016; time 2 × age: β = 8.7e–5, p = 0.043, d = 0.17). Model comparisons supported the quadratic over the linear model for WMV (AIC: 630.93 vs. 633.88; ΔAIC = 2.95) and medial temporal PCr/ATP (AIC: − 27.70 vs. − 24.75; ΔAIC = 2.95), but not for middle frontal PCr/ATP. Consistent with limited power for nonlinear effects, these findings are considered descriptive and hypothesis-generating, pending replication in larger samples. No significant effects were found for regional volumes or CBF measures (Table 4 ). Table 4 Age at surgery effects on biomarker changes. Biomarker Region Linear Non-linear Estimate Std. error P -value Estimate Std. error P -value GMV Hippocampus − 0.082 0.176 0.646 − 0.016 0.013 0.225 Hypothalamus − 0.020 0.026 0.457 0.001 0.002 0.496 Parahippocampal gyrus 0.045 0.240 0.855 − 0.002 0.007 0.765 Global white matter − 106.20 37.69 0.024 2.509 1.073 0.049 CBF Medial temporal 0.004 0.009 0.695 0.001 0.001 0.497 Middle frontal − 0.004 0.016 0.801 0.001 0.001 0.588 Superior frontal − 0.008 0.016 0.614 0.001 0.001 0.757 PCr/ATP Medial temporal − 0.002 0.001 0.013 0.001 0.001 0.017 Middle frontal − 0.003 0.001 0.016 0.001 0.001 0.043 Superior frontal − 0.002 0.001 0.111 0.001 0.001 0.149 Results from mixed effects models testing for age × group effects on brain biomarker outcomes in the premenopausal oophorectomy (PO) group. Significant results are in bold. CBF, cerebral blood flow; PCr/ATP, phosphocreatine to adenosine triphosphate ratios. Age at surgery effects on biomarker changes. Results from mixed effects models testing for age × group effects on brain biomarker outcomes in the premenopausal oophorectomy (PO) group. Significant results are in bold. CBF, cerebral blood flow; PCr/ATP, phosphocreatine to adenosine triphosphate ratios. A younger age at PO was associated with a steeper rise in LH post-surgery (time × age: β  = − 0.004, p  = 0.035) and with a marginal steeper decline in estradiol (time × age: β  = − 0.004, p  = 0.078), whereas no associations were found with FSH (time × age: β  = − 0.002, p  = 0.218). No significant group × time interactions were observed for cognitive outcomes (Supplementary Table S3 ). No time × age at surgery interactions emerged in either linear or non-linear models (P’s > 0.21). Therefore, associations with brain biomarkers were not examined.

Hysterectomy is the second most common surgery among women in the United States, and the most commonly performed procedure among women aged 40–69 years 1 , with over 500,000 hysterectomies performed annually 1 – 3 . Elective bilateral salpingo-oophorectomy (BSO, the surgical removal of both ovaries and fallopian tubes) was and often still is routinely offered to women at the time of hysterectomy as a prophylactic procedure to prevent ovarian cancer 1 – 3 . An estimated 23–40% of women aged 40–44 years, and 45–63% of women aged 45–49 years, and 78% of women aged 50–54 undergo BSO at the time of hysterectomy 4 – 7 . Premenopausal bilateral oophorectomy (PO) causes abrupt endocrine dysfunction (e.g., decline of estrogen, progesterone, testosterone, and an increase in gonadotropins) and disruption of the hypothalamic-pituitary-gonadal (HPG) axis 8 – 10 . PO has been linked with accelerated biological aging 10 – 12 and accumulation of multimorbidity, with heightened risk of age-related brain disorders such as depression, stroke, parkinsonism, and dementia 13 – 15 . Whether PO directly leads to these risks is unclear, as no studies have investigated the impact of the procedure on the brain by comparing data collected before and after surgery. In neuroimaging studies, women with a history of PO exhibit lower medial temporal lobe volume 16 – 18 , reduced white matter integrity 16 , 19 , altered functional brain phenotypes 20 , and greater amyloid-beta (Aβ) burden 21 , 22 compared to spontaneous, later-onset menopause. However, all prior studies analyzed data acquired years to decades after surgery 16 – 22 , and most included PO patients using menopause hormone therapy (MHT) 16 – 21 , which precludes determining whether the observed brain abnormalities are attributable to the procedure itself or influenced by exogenous hormone use. Other studies reported negative associations between early or premature menopause—primarily, but not exclusively due to oophorectomy—with gray (GMV) and white matter volume (WMV) declines 23 – 26 , Aβ 21 and tau pathology 27 in later life. Herein, we conducted a prospective, pre- vs. post-surgery multimodality Magnetic Resonance (MR) imaging study with a matched control design in women undergoing oophorectomy for benign or prophylactic indications before menopause, MHT non-users. This was an exploratory, hypothesis-generating mechanistic study designed to establish feasibility and generate preliminary estimates for future studies. We examined the impact of PO status and age at surgery on GMV and WMV, cerebral blood flow (CBF) using Arterial Spin Labeling (ASL)-MR, and brain mitochondrial function using 31 Phosphorus MR Spectroscopy ( 31 P-MRS). In exploratory analyses, we examined associations of brain biomarker changes with plasma hormone levels and cognition.

Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2