Abstract
Reductions in the ability to encode and retrieve past experiences in rich spatial contextual detail (episodic memory) are apparent by midlife—a time when most females experience spontaneous menopause. Yet, little is known about how menopause status affects episodic memory-related brain activity at encoding and retrieval in middle-aged premenopausal and postmenopausal females, and whether any observed group differences in brain activity and memory performance correlate with chronological age within group. We conducted an event-related task fMRI study of episodic memory for spatial context to address this knowledge gap. Multivariate behavioral partial least squares was used to investigate how chronological age and retrieval accuracy correlated with brain activity in 31 premenopausal females (age range, 39.55–53.30 years; mean age, 44.28 years; SD age, 3.12 years) and 41 postmenopausal females (age range, 46.70–65.14 years; mean age, 57.56 years; SD age, 3.93 years). We found that postmenopausal status, and advanced age within postmenopause, was associated with lower spatial context memory. The fMRI analysis showed that only in postmenopausal females, advanced age was correlated with altered activity in occipitotemporal and parahippocampal cortices during encoding and retrieval, and poorer spatial context memory performance. In contrast, only premenopausal females exhibited an overlap in encoding and retrieval activity in angular gyrus/inferior parietal cortex, midline cortical regions, and prefrontal cortex, which correlated with better spatial context retrieval accuracy. These results highlight how menopause status and chronological age, nested within menopause group, affect episodic memory and its neural correlates at midlife.
Significance Statement
This is the first fMRI study to examine how premenopause and postmenopause status affect the neural correlates of episodic memory encoding and retrieval, and how chronological age contributes to any observed group similarities and differences. We found that both menopause status (endocrine age) and chronological age affect spatial context memory and its neural correlates. Menopause status directly affected the direction of age-related and performance-related correlations with brain activity in parahippocampal and occipitotemporal cortices across encoding and retrieval. Moreover, we found that only premenopausal females exhibited cortical reinstatement of encoding-related activity in midline cortical, prefrontal, and angular gyrus/inferior parietal cortex, at retrieval. This suggests that spatial context memory abilities may rely on distinct brain systems at premenopause compared with postmenopause.
Introduction
Aging is associated with declines in the ability to encode and consciously recollect past events in rich contextual detail (episodic memory; Tulving et al., 1972) and related brain activation differences in the medial temporal lobe, occipitotemporal cortex, prefrontal cortex (PFC), inferior parietal cortex, and midline cortical regions (Persson et al., 2012; Spaniol and Grady, 2012; Maillet and Rajah, 2013, 2014b; Rajah et al., 2015). Importantly, there is growing evidence that age-related episodic memory decline, as measured by object-location spatial context memory tasks, arises at midlife (Cansino et al., 2012) and is associated with altered PFC and occipitotemporal activity (Kwon et al., 2016). Midlife is also the time that most females with ovaries experience natural/spontaneous menopause and transition from being in premenopause (Pre-Meno) to postmenopause (Post-Meno; Harlow et al., 2012). Some postmenopausal females report cognitive issues, including brain fog and forgetfulness (Greendale et al., 2020). This raises the possibility that some of the prior reports of chronological age effects on memory and brain function at midlife may have been driven in part by postmenopausal females. Therefore, there is a need to bridge results from the cognitive neuroscience of aging and memory field with those from the menopause and endocrine aging field to advance our understanding of how menopause status affects episodic memory and related brain function at midlife, and if any observed effects correlate with the chronological age of females at postmenopause, compared with premenopause (Taylor et al., 2019).
However, to our knowledge, only one event-related task functional magnetic resonance imaging (fMRI) study has examined the effect of menopause status on brain activity during the performance of a verbal episodic encoding task (Jacobs et al., 2016). They found that postmenopausal status was associated with less activity in the hippocampus compared to premenopausal status. However, there was no behavioral effect of menopause status on verbal episodic retrieval, suggesting that the observed effects may not reflect the neural basis of memory disturbances experienced by some females at menopause (Greendale et al., 2020). Also, in this study the chronological age of premenopausal and postmenopausal females was matched to control for age effects. Thus, it remains unclear whether the observed menopause-related differences in brain activity differ with chronological age within premenopause and/or postmenopause groups. Given that past work has suggested that brain and cognitive dysfunction at menopause is transient (Greendale et al., 2009), it is important to clarify whether and how chronological age affects memory and brain differences at premenopause, compared to postmenopause. Such information can help determine the following: (1) what type of intervention may benefit females experiencing memory decline—ones that target endocrine function at menopause (e.g., hormone replacement therapies), if effects are stable, or ones that target both endocrine function and senescence, if effects vary with age at postmenopause (Doty et al., 2015); and (2) the timing and duration of these interventions. Finally, no study, to our knowledge, has examined how menopause status affects episodic retrieval-related brain activity, which highlights the paucity of knowledge about midlife brain health and episodic memory function in cognitively unimpaired females.
In the current task fMRI study, we investigate whether and how menopause status affects episodic memory and related brain function during encoding and retrieval in a broad sample of middle-aged premenopausal and postmenopausal females. We use a face-location spatial context task fMRI paradigm that can detect episodic memory decline at midlife (Kwon et al., 2016). Performance on such tasks is correlated with the structural integrity and function of hippocampus (Beer et al., 2018; Snytte et al., 2020). Thus, we hypothesized that this paradigm would be sensitive for detecting postmenopause declines in episodic memory and associated differences in brain function. We also examine how chronological age, nested within menopause status, correlates with memory-related brain activity to determine whether chronological aging contributes to any observed group similarities and/or differences. We hypothesize that there will be differences in chronological age effects on hippocampus, PFC, and occipitotemporal cortex activity at encoding and retrieval based on menopause status.
Materials and Methods
Participants
Ninety-six middle-aged (age range, 39.5–65 years) females with ovaries participated in this study. Participants were categorized as being Pre-Meno or Post-Meno based on the Stages of Reproductive Aging Workshop + 10 (STRAW + 10) criteria (Harlow et al., 2012). Based on these criteria there were 33 Pre-Meno participants (age range, 39.55–53.30 years; mean age, 44.41; SD, age, 3.12; mean education, 16.44 years; ethnicity: white, 81.8%; Latin American, 9.1%; Black, 3.0%; Chinese, 3.0%; South Asian, 3.0%) and 63 Post-Meno participants (age range, 46.70–65.14 years; mean age, 57.05 years; SD age, 3.93; mean education, 15.39 years; ethnicity: White, 87.3%; Black, 3.2%; Indigenous, 3.2%; Chinese, 1.6%; South Asian, 1.6%; Middle Eastern and North African, 3.2%). This study was approved by the research ethics board of the Montreal West Island Integrated University Health and Social Services Center—subcommittee for mental health and neuroscience, and all participants signed informed consent forms.
Behavioral methods
Enrollment
Participants were recruited via advertisements posted on websites and around the Montreal community. Participants first signed an online consent form and then proceeded to fill out an online questionnaire that collected demographics, medical history, reproductive history, and education information. This information was used to screen participants for inclusion/exclusion in the study. Individuals who met the inclusion/exclusion criteria listed below were invited to Behavioral Session 1 at the Cerebral Imaging Center (CIC) at the Douglas Mental Health University Institute (DMHUI).
Inclusion criteria
The minimum criteria were high school education, willingness to provide a blood sample, and in general good health. Participants were excluded if they reported any of the following in their medical history: current use of hormone replacement therapy (HRT), bilateral oophorectomy, untreated cataract/glaucoma/age-related maculopathy, uncontrolled hypertension, untreated high cholesterol, diabetes, history of estrogen-related cancers, chemotherapy, neurologic diseases or history of serious head injury, history of major psychiatric disorders, claustrophobia, history of a substance use disorder, or currently smoking >40 cigarettes/d. Participants were also excluded if they did not meet the requirements for MRI safety (e.g., implanted pacemaker). Females who were pregnant, perimenopausal, or had indeterminate menopausal status, based on self-report and hormone levels, were excluded from the study.
Behavioral session
After obtaining consent, participants filled out the MRI safety questionnaire. Next, they completed a battery of standardized psychiatric and neuropsychological assessments, which included the following: the Mini-International Neuropsychiatric Interview (Sheehan et al., 1998) exclusion criteria yielding indications of undiagnosed psychiatric illness; the Beck Depression Inventory II (BDI-II; Beck et al., 1997) inclusion cutoff of ≤19; and the Mini-Mental State Examination (Folstein et al., 1975) inclusion cutoff of ≥27. Participants then donated blood samples for hormonal assays to confirm self-reported menopausal status and performed a practice version of the spatial context memory task in a mock MRI scanner to familiarize them with the task and ensure they felt comfortable enough to partake in the real MRI scan. Participants who met the neuropsychological inclusion criteria stated above, were Pre-Meno or Post-Meno based on the STRAW + 10 criteria and hormone levels (Harlow et al., 2012), and were able to perform the practice fMRI task were invited to a second visit to undergo MRI scans at the CIC at DMHUI.
Endocrine assessments
Estradiol-17β (E2), follicle stimulating hormone (FSH), luteinizing hormone (LH), and progesterone (P) levels were assessed to corroborate menopausal staging based on menstrual cycling and STRAW + 10. Endocrine measurements were assessed in plasma, derived from heparin blood collection tubes. Blood was drawn on nonfasting individuals by a certified research nurse at the DMHUI during session 2 (MRI visit). Specimens were analyzed at the McGill University Health Center (Glen site) Clinical Laboratories in Montreal. Endocrine chemiluminescent immunoassays were performed on an Access Immunoassay System (Beckman Coulter), using the company's reagents. All participants were categorized as Pre-Meno or Post-Meno based on the self-report and measured hormone for FSH as outlined by STRAW + 10 criteria. Females who were perimenopausal or whose menopausal status was indeterminate were excluded from the study.
Task fMRI session
Upon arrival at the CIC, participants first took a pregnancy test. Participants who were not pregnant were then scanned while performing the spatial context memory task. Participants were asked to lie in a supine position in a 3 T scanner (Prisma-Fit, Siemens), wearing a standard 32-channel head coil. High-resolution T1-weighted (T1w) anatomic images were collected using a 3D magnetization-prepared rapid acquisition gradient echo sequence [repetition time (TR), 2300 ms; echo time (TE), 2.36 ms; flip angle, 9°; 192 1-mm-thick transverse slices; 1 × 1 × 1 mm voxels; field-of-view (FOV), 256 mm2; acquisition time, 5 min, 3 s ] to use in the registration and normalization of the fMRI data. While performing the spatial context memory task, functional BOLD images were acquired using single shot T2*-weighted gradient echoplanar imaging (EPI) pulse sequence (TR, 2000 ms; TE, 30 ms; FOV, 256 mm2; matrix size, 64 × 64; in-plane resolution, 4 × 4 mm).
A mixed rapid event-related design was used for the task fMRI portion of the experiment (Dale, 1999). Participants were scanned during encoding and retrieval phases of Easy and Hard spatial context memory tasks. Task Difficulty was manipulated by increasing encoding load (details below). A participant could perform up to four Easy and four Hard task runs (maximum, eight runs total). Easy runs included two Easy spatial context memory tasks (total run duration, 9 min, 42 s), whereas Hard runs only consisted of one experimental block (total run duration, 7 min, 14 s). Run order was counterbalanced across all participants. The total length of the task fMRI session was ∼1 h, 7 min, and 44 s.
The tasks were presented through E-Prime software (Psychology Software Tools), back-projected into the scanner bore, and made visible to participants via a mirror mounted within the head coil. Participants were provided with two four-button inline fiber optic response boxes to respond throughout the task. During each run, response options and the corresponding buttons for each task were displayed at the bottom of the screen for clarity (Fig. 1, visual summary of the task fMRI protocol). Response accuracy and reaction times (RTs; in milliseconds) were collected for all responses by E-Prime software.
Task fMRI paradigm design. Research participants were asked to encode face–location associations and make a pleasant/neutral decision for face stimuli presented in one of four quadrants. There was a 1 min interval between encoding and retrieval. During retrieval, participants were presented with an equal number of previously encoded (“old”) and novel face stimuli in the center of the screen and asked to make a six-alternative forced-choice decision. Details about the paradigm are presented in the Materials and Methods section.
Encoding phase
Participants were presented with instructions for the first 6 s of each encoding phase. Participants were informed that they would see a series of black and white photographs of faces that varied in age (from toddlers to older adults). They were instructed to memorize each face and its spatial location and to rate each face as either pleasant or neutral by pressing the corresponding button on the provided response box. The pleasantness rating was added to ensure participants deeply encoded the stimuli (Bernstein et al., 2002). The stimulus set has been described in detail in prior publications (Rajah et al., 2008). There were 50% girl/women and 50% boy/men faces presented, and the race and ethnicity of the faces presented were 83% White, 8.5% Black, and 8.5% Other.
Following the instructions, participants were presented sequentially with either 6 or 12 face stimuli. Each face stimulus was presented one at a time (2 s/stimulus) in one of four quadrants on the computer screen (Fig. 1). The intertrial interval (ITI) was varied (ITI range, 2.5–7.5 s; mean ITI, 4.17 s) to add jitter to fMRI data collection (Dale and Buckner, 1997). Conditions in which 6 faces were presented at encoding were considered the “Easy,” and those in which 12 faces were presented at encoding were considered “Hard.” Thus, the experiment included a Task Difficulty manipulation, which was based on the encoding load. This allowed us to assess within-group and between-group differences in performance as a function of increased task demands. Participants saw up to 48 faces during the Easy encoding condition, and up to 48 faces presented during the Hard encoding condition at the end of four Easy runs. However, because of participants withdrawing participation and software errors, not all participants experienced all runs during the task fMRI session. The minimum number of Easy runs experienced by any participant was two. Each Easy run had two tasks, so the minimum number of Easy tasks experienced was four. Thus, the minimum number of faces encoded under the Easy condition was calculated as 4 tasks × 6 encoding stimuli = 24 faces. The minimum number of Hard runs experienced by any participant was three. Each Hard run had one Hard task. Thus, the minimum number of Hard encoding events experienced by a participant was 36. At the end of the encoding phase, participants rated their confidence on how well they performed the task on a 4-point scale from very poorly (1) to very well (4) during a 60 s break.
Retrieval phase
Participants were presented with instructions for 14 s. They were informed that they would be shown a series of previously encoded (old) and novel (new) photographs of faces, presented to them one at a time. They were instructed to respond to each face by pressing one of six buttons corresponding to the following retrieval responses: (1) N, the face is new; (2) F, the face is familiar but I do not remember its location; (3) TL, I remember the face, and it was previously on the top left; (4) BL, I remember the face, and it was previously on the bottom left; (5) TR, I remember the face, and it was previously on the top right; and (6) BR, I remember the face, and it was previously on the bottom right. Participants were instructed not to guess and only respond with buttons 3 to 6 if they clearly recollected the face and its location.
Following the instructions, participants saw a series of either 12 (Easy condition) or 24 (Hard condition) faces, presented 1 at a time, per task. Each face was presented for 4 s in the center of the screen. There was a variable ITI (ITI range, 2.5–7.5 s; mean ITI, 4.17 s). A participant saw up to a total of 96 faces (48 old, 48 new) at retrieval following four Easy and four Hard runs per condition, respectively. The minimum number of Easy runs experienced by any participant was two. Each Easy run had two tasks, so the minimum number of Easy tasks experienced was four. Thus, the minimum number of retrieval events under the Easy condition was calculated as 4 tasks × 12 encoding stimuli = 48 retrieval events (24 old, 24 new). The minimum number of Hard runs experienced by any participant was three. Each Hard run had one Hard task. Thus, the minimum number of Hard encoding events experienced by a participant was as follows: 3 × 24 = 72 (36 old, 36 new). At the end of the retrieval phase participants were given 60 s to rate how well they thought they performed on a 4-point scale from very poorly (1) to very well (4).
Data analysis: behavioral analyses
Mean accuracy (percentage correct) and reaction time (in milliseconds) were calculated for each participant for the following retrieval response types within Easy and Hard tasks:
Correct Spatial Context Retrieval (CS) Easy and Hard: total number of correct associative retrieval trials (i.e., three, four, five, and six button presses made for previously seen faces; Fig. 1) divided by the total number of previously seen faces presented at retrieval.
Correct Recognition (Recog) Easy and Hard: total number of correct retrieval trials for recognition without memory for location (i.e., two responses for previously seen faces) divided by the total number of previously seen faces presented at retrieval.
Partial least-squares regression for accuracy and RT
To examine the contributions of Menopause Group and Age (years) to variability in Recog and CS on Easy and Hard versions of tasks, as well as RT values for these judgments, we conducted partial least-squares (PLS) regression analyses. Specifically, we ran multiresponse PLS regressions in R (version 4.1.3) using the plsreg2 function from the plsdepot package [version 0.2.0; available at: https://CRAN.R-project.org/package=plsdepot]. We selected this analytical method as it is capable of handling multicollinearity in predictor variables (as is the case for Menopause Group and Age) by creating a set of uncorrelated components to use in the model, and it can facilitate the inclusion of multiple response variables in one model (Wold et al., 2001; Abdi, 2010). The aim of PLS regression is to predict Y (i.e., the outcome variables) from X (i.e., the predictor variables) and to describe their common structure. To achieve this, X and Y are first decomposed to form a set of latent variables (LVs) that maximally explain the shared variance (i.e., the covariance) between the original variables. The ability of a given LV to predict Y is then typically assessed via cross-validation, a process that produces Q2 values. LVs were considered significant and retained if Q2 ≥ 0.0975 (Abdi, 2010). For all such LVs, the variance explained (R2) in both X and Y is reported. Although PLS regression is capable of handling cases involving highly collinear predictor variables, this type of analysis does not provide insight as to whether age is differentially related to spatial context retrieval accuracy at specific stages of menopause. That is, this method does not allow us to examine within-group age effects. To address this, we conducted linear mixed-effects models for Pre-Meno and Post-Meno females in R using the lme4 package (version 1.1-29; Bates et al., 2015). Task Difficulty (Easy, Hard) and Age (years) were entered as fixed effects, as was their interaction. Task Difficulty was coded using deviation coding, and Age was standardized. Both models included a random by-participant intercept. Statistical significance of fixed effects was determined via Satterthwaite approximations, implemented by the lmerTest package (version 3.1-3; Kuznetsova et al., 2017).
To assess whether effects of age provided a better fitting model, nested hierarchical linear mixed-effect models were fitted (with and without Age as a predictor) and were then compared on values assessing goodness-of-fit (χ2) and model parsimony [Akaike Information Criterion (AIC); ΔAIC ≥ 2 as the threshold for model selection].
Task fMRI analyses
Image preprocessing
The DICOM files were first converted to NIFTI format and organized using the Brain Imaging Data Structure (Gorgolewski et al., 2016). Volumes collected during the first 10 s of scanning were removed to ensure that magnetization had stabilized. Preprocessing was then conducted using fMRIPrep version 20.2 (Esteban et al., 2019). FMRIPrep is a robust preprocessing workflow in Python version 3.0 that implements tools from various software packages including Advanced Normalization Tools version 2.3.3 (ANTs), FMRIB Software Library version 5.0.9 (FSL), and FreeSurfer version 6.0.1. The following preprocessing description was generated by fMRIPrep (Esteban et al., 2019):
For each of the eight BOLD runs found per participant (across all tasks and sessions), the following preprocessing was performed. First, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep. A B0-nonuniformity map (or fieldmap) was estimated based on a phase-difference map calculated with a dual-echo gradient-recalled echo sequence, processed with a custom workflow of SDCFlows inspired by the epidewarp.fsl script with further improvements obtained using HCP Pipelines (Glasser et al., 2013). The fieldmap was then coregistered to the target EPI reference run and converted to a displacements field map (amenable to registration tools; e.g., ANTs) with FSL fugue and other SDCflows tools. Based on the estimated susceptibility distortion, a corrected EPI reference was calculated for a more accurate coregistration with the anatomic reference. The BOLD reference was then coregistered to the T1w reference using bbregister (FreeSurfer), which implements boundary-based registration (Greve and Fischl, 2009). Coregistration was configured with 6 df. Head motion parameters with respect to the BOLD reference (transformation matrices, and six corresponding rotation and translation parameters) are estimated before any spatiotemporal filtering using mcflirt (FSL version 5.0.9; Jenkinson et al., 2002). The BOLD time series were resampled onto their original, native space by applying a single, composite transform to correct for head motion and susceptibility distortions. These resampled BOLD time series will be referred to as preprocessed BOLD in original space, or just preprocessed BOLD. The BOLD time series were resampled into standard space, generating a preprocessed BOLD run in MNI152NLin2009cAsym space. First, a reference volume and its skull-stripped version were generated using a custom methodology of fMRIPrep. Several confounding time series were calculated based on the preprocessed BOLD: framewise displacement (FD), DVARS, and three region-wise global signals. FD was computed using two formulations following Power et al. (2014; absolute sum of relative motions) and Jenkinson et al. (2002; relative root mean square displacement between affines). FD and DVARS are calculated for each functional run, both using their implementations in Nipype (following the definitions of Power et al., 2014). Three global signals are extracted within the CSF, the white matter (WM), and the whole-brain masks. Additionally, a set of physiological regressors were extracted to allow for component-based noise correction (CompCor; Behzadi et al., 2007).
Principal components are estimated after high-pass filtering the preprocessed BOLD time series (using a discrete cosine filter with 128 s cutoff) for the two CompCor variants: temporal (tCompCor) and anatomic (aCompCor). tCompCor components are then calculated from the top 2% variable voxels within the brain mask. For aCompCor, three probabilistic masks (CSF, WM, and combined CSF + WM) are generated in anatomic space. The implementation differs from that of Behzadi et al. (2007) in that instead of eroding the masks by 2 pixels on BOLD space, the aCompCor masks are subtracted from a mask of pixels that likely contain a volume fraction of gray matter (GM). This mask is obtained by dilating a GM mask extracted from the FreeSurfer aseg segmentation, and it ensures that components are not extracted from voxels containing a minimal fraction of GM. Finally, these masks are resampled into BOLD space and binarized by thresholding at 0.99 (as in the original implementation). Components are also calculated separately within the WM and CSF masks. For each CompCor decomposition, the k components with the largest singular values are retained, such that the time series of the retained components are sufficient to explain 50% of the variance across the nuisance mask (CSF, WM, combined, or temporal). The remaining components are dropped from consideration. The head motion estimates calculated in the correction step were also placed within the corresponding confounds file. The confound time series derived from head motion estimates and global signals were expanded with the inclusion of temporal derivatives and quadratic terms for each (Satterthwaite et al., 2013). Frames that exceeded a threshold of 0.5 mm FD or 1.5 standardized DVARS were annotated as motion outliers. All resamplings can be performed with a single interpolation step by composing all the pertinent transformations (i.e., head motion transform matrices, susceptibility distortion correction, when available, and coregistrations to anatomic and output spaces). Gridded (volumetric) resamplings were performed using antsApplyTransforms, configured with Lanczos interpolation to minimize the smoothing effects of other kernels. Nongridded (surface) resamplings were performed using mri_vol2surf (FreeSurfer).
In addition to fMRIPrep, further preprocessing steps were conducted using custom code in Python 3.0 and Nilearn libraries (Abraham et al., 2014). If a participant had >15% of all their scans showing motion >1 mm FD, they were removed from the analysis. For participants with motion <15% across all scans, the normalized scans from fMRIPrep were scrubbed for motion artifacts as follows: (1) when <15% of scans across all runs were above the 1 mm cutoff, and only one or two consecutive volumes exceeded this cutoff, the volume was replaced by the average of the previous and the subsequent volumes in time [e.g., a scan T_i that had FD > 1 mm was replaced by the mean/average of scans T_(i – 1) and T_(i + 1)]; and (2) when <15% of scans across all runs exceeded 1 mm FD and there were three or more consecutive volumes that exceeded 1 mm FD, then onsets including those volumes were excluded from the PLS analysis. The scans were then smoothed using a Gaussian filter (FWHM = 8 mm). Finally, confounds such as WM, CSF, and six motion parameters were regressed out from all the runs. The average percentage of total scans scrubbed postprocessing was –0.52% (1.04). The maximum percentage of scans scrubbed within a participant was 6.68%. The average number of onsets removed/censored was –0.35 (1.08). There were only nine participants for whom onsets were removed. A maximum of six onsets was excluded for a given participant.
Behavior PLS analysis
A multivariate between-group behavior PLS (B-PLS; McIntosh and Lobaugh, 2004) analysis was conducted via open source PLSGUI software (https://www.rotman-baycrest.on.ca/index.php?section=345) in MATLAB (version R2022a; MathWorks). Partial least-squares analysis was chosen because it is a powerful method that can be used to objectively assess spatiotemporal functional patterns in neuroimaging datasets. Moreover, between-group B-PLS directly assesses the correlation between brain activity and different demographic and/or behavioral measures within group (i.e., age and memory performance; McIntosh et al., 2004). When conducting a PLS analysis, the fMRI and behavioral data are stored in two separate matrices. The fMRI data matrix was organized with rows representing brain activity for each event type of experimental interest nested within participants, and participants nested within groups of Pre-Meno females and Post-Meno females. In the current analysis, the rows included activity for the following four event types of interest nested within participant and nested within group: all encoding events during Easy conditions (EncEasy); all encoding events during Hard conditions (EncHard); all retrieval events during Easy conditions (RetEasy); and all retrieval events during Hard conditions (RetHard). The columns of the fMRI data matrix contain mean-centered BOLD activity of all brain voxels at event onset and the subsequent seven scans/lags, wherein each lag is equivalent to the TR of the T2*-weighted gradient EPI sequence or 2000 ms, thus capturing 16 s of whole-brain activity following every event onset. PLS averages the number of events for each event type/condition by lag. As stated above, the minimum number of encoding events experienced by a participant during Easy conditions was 24, and during Hard condition it was 36. The minimum number of Easy retrieval events experienced by a participant during Easy conditions was 48 (24 old, 24 new), and during Hard conditions it was 72 (36 old, 36 new).
The behavioral data were stored in a separate matrix. The rows of the behavior matrix are stacked in the same order as the fMRI data matrix The columns of the behavioral matrix included vectors of participants' behavioral data for the following: (1) standardized age, (2) standardized mean spatial context accuracy, and (3) standardized mean recognition accuracy. The behavior matrix was cross-correlated with the fMRI data matrix and underwent singular value decomposition (SVD). SVD yields a series of orthogonal LVs that is equal to the number of groups-by-event type per condition (2 groups × 4 event type/conditions = 8 LVs). Each LV is composed of a singular value, a singular image, and a correlation profile. The singular value represents the proportion of covariance accounted for by the LV. The singular image includes positive and/or negative saliences that establish the weighted contributions of each voxel at a given time lag, producing a spatiotemporal pattern of whole-brain activity. Time lags are the number of TRs after the stimulus of interest is presented. The correlation profile portrays the association between the pattern of brain activity from the singular image with the behavioral measures. In interpreting the correlation profiles, it is important to note that PLS results are symmetrical, such that, negative correlations between a behavioral vector (age, performance) and a negative salience brain regions are symmetrical with more activity in these negative salience regions being positively correlated with these same behavioral vectors. The opposite symmetry can be stated for positive correlations with positive salience regions.
Permutation testing was conducted on the LVs to establish significance (p < 0.01 and cross-block covariance > 10%; McIntosh et al., 2004). The permutation test involved sampling without replacement to reassign links between subjects' behavioral vector measures and event/condition within subject. For each permuted iteration, a PLS was recalculated, and the probability that the permuted singular values exceeded the observed singular value for the original LV was used to assess significance at p < 0.05 (McIntosh et al., 2004). To identify stable voxels that consistently contributed to the correlation profile within each LV, the SEs of the voxel saliences for each LV were estimated via 500 bootstraps, sampling subjects with replacement while maintaining the order of event types for all subjects. For each voxel, a value similar to a z score known as the bootstrap ratio (BSR) was computed, reflecting the ratio of the original voxel salience to the estimated SE for that voxel. Voxels with BSR values of ± 3.28 (equivalent to p < 0.001) and a minimum spatial extent = 10 contiguous voxels, were retained and highlighted in the singular image. BSR values reflect the stability of voxel saliences. A voxel salience whose value is dependent on the observations in the sample is less precise than one that remains stable regardless of the samples chosen (McIntosh and Lobaugh, 2004). The bootstrapping results were used to calculate 95% confidence intervals (CIs) for the correlations between event-related brain activity and the three following behavioral vectors: chronological age, spatial context retrieval accuracy, and recognition accuracy during EncEasy, EncHard, RetEasy, RetHard conditions. If the CI crosses 0, the effect is not significant for that group and condition. In this B-PLS analysis, a group difference in the correlation between chronological age and memory-related brain activity would manifest as either the correlations between age and memory-related brain activity being in opposite directions between Pre-Meno and Post-Meno or as one group showing a correlation between age and memory-related brain activity and not the other. Group similarities in age differences in memory-related brain activity would present as both groups exhibiting the same direction of correlation between age and brain activity.
Temporal brain scores were calculated to obtain time lags with the strongest correlation profile for significant LVs from each PLS analysis. Temporal brain scores reveal the strength of the pattern of brain activity for each participant at each time lag. We retained only peak coordinates from time lags 3–5, which exhibited maximal effects related to the PLS effect. Identified peaks were converted to Talairach space using the icbm2tal transform (Lancaster et al., 2007), implemented in GingerAle 2.3 (Eickhoff et al., 2009). Relevant Brodmann areas (BAs) were established using the Talairach and Tournoux (1988) atlas and confirmed in FSL. Peak coordinates from the cerebellum and brainstem were not included as the fMRI acquisition in these regions was incomplete.
Post hoc analyses
Post hoc between-group, repeated-measures multivariate ANOVAs (MANOVAs) were conducted for each significant LV to determine whether there were significant group differences in LV brain scores based on a female's current antidepressant use, hormonal birth control (HBC) use, and diagnosis of polycystic ovary syndrome (PCOS). Significance was assessed at p < 0.05 corrected.
Data availability
Code (MATLAB, R) used for the present study is made publicly available on our Lab GitHub page (https://github.com/RajahLab/BHAMM_TaskfMRI_scripts). Readers seeking access to data should e-mail the Principal Investigator of the Brain Health at Midlife and Menopause (BHAMM) Study (M.N.R.; natasha.rajah{at}torontomu.ca) for further information.
Results
Behavioral results
A total of 24 participants were excluded from behavioral and fMRI analyses, as follows: 11 Post-Meno participants were excluded for being on HRT (9 participants) or HBC (2 participants) for hormonal regulation, 1 participant because of incidental finding, 1 participant because of scanner technical issues, 2 participants because of excessive movement (>15% of all scans with FD > 1 mm) during MRI scanning, 8 participants for being behavioral outliers and/or task noncompliant, and 1 participant was identified as a PLS brain scores behavior correlation outlier. A behavioral outlier was defined as an individual that had a Cook's D value, which was 3 SDs from the group mean on three or more response types. A noncompliant responder was defined as someone who performed poorer than chance on the correct rejections and/or had zero spatial context retrieval responses.
Demographic and fMRI behavioral data separated by reproductive stage for the final sample (N = 72) are reported in Table 1. The age range for the 31 Pre-Meno females was 39.55–53.30 years, with an education range of 12–20 years. The age range for the 41 Post-Meno females was 52.30–65.14 years, with an education range of 11–20 years. χ2 tests were conducted to determine whether there were group differences in the proportion of antidepressant medication use, use of HBC (i.e., oral contraceptive, patch, intrauterine devices), or diagnosis of PCOS. There was a significant group difference in HBC use (p = 0.018), with Pre-Meno females having 12.0% HBC users for (n = 4) and Post-Meno females having 0%. We did not exclude the four Pre-Meno HBC users as their reproductive health data indicated it was being used for contraceptive purposes, and they were not identified as outliers in the behavioral or fMRI analyses. One-way ANOVAs were conducted to compare Pre-Meno and Post-Meno females on demographic variables (age, education, handedness, BMI, BDI-II scores), blood-based measures of sex hormone levels (E2, FSH, LH, P), and task fMRI behavioral data. There was a significant group difference in age and hormone levels. Post-Meno females were significantly older than Pre-Meno females. Post-Meno females had significantly higher FSH and LH levels and lower E2 and P levels compared with Pre-Meno females, consistent with their menopause staging. Detailed analysis of the task fMRI analysis data are presented below.
Demographic information and fMRI behavioral measures by reproductive group
Behavioral analysis of task fMRI accuracy and RT data
The PLS regression analysis of accuracy data identified one LV with a Q2 value above threshold (Q2 = 0.1572). This LV accounted for 94.47% of the variance in X (i.e., Menopause Group, Age) and 17.81% of the variance in Y (i.e., CS Easy/Hard, Recog Easy/Hard). Figure 2A highlights the correlations between the original predictor/outcome variables and the LV. For the predictor variables, both Menopause Group and Age were negatively associated with the LV. For the outcome variables, Recog Easy and Recog Hard were negatively correlated with the LV (albeit relatively weakly), whereas CS Easy and CS Hard were positively correlated with the LV. These results indicate that the LV captured a pattern in which Menopause Group and Age were positively correlated with Recog but negatively correlated with CS, for both Easy and Hard conditions. This indicates that both predictor variables were associated with changes in spatial context memory performance. By contrast, the PLS regression analysis of RT data failed to identify an LV with a Q2 value >0.0975 (Abdi, 2010).
Behavioral results. A, The correlation profile for the LV identified by PLS regression. The bars reflect the correlation between the original independent (blue)/dependent (orange) variables and the LV. B, Association between age (standardized) and spatial context retrieval accuracy as a function of Task Difficulty (Easy, Hard) in Pre-Meno females (left) and Post-Meno females (right). Lines of best fit and 95% confidence intervals are based on predicted values.
As a follow-up to this analysis, we conducted linear mixed-effects models to test the effect of Age and Task Difficulty (Easy, Hard) on CS within each group. For Pre-Meno females, we observed a statistically significant effect of Task Difficulty (F(1,29) = 4.905, p = 0.035) but not Age (F(1,29) = 1.275, p = 0.268). The effect of Task Difficulty was driven by lower CS on the Hard compared with the Easy task version. Moreover, there was no statistically significant interaction (F(1,29) = 1.424, p = 0.242). For Post-Meno females, we observed statistically significant effects of Task Difficulty (F(1,39) = 20.26, p < 0.001) and Age (F(1,39) = 11.372, p = 0.002). These effects were driven by lower CS on the Hard task compared with the Easy task version and with advanced age. Notably, no interaction effect was present (F(1,39) = 0.125, p = 0.726). An overview is presented in Figure 2B. These results suggest that the effect of chronological age on spatial context retrieval accuracy is dependent on menopausal status, reflecting a more complex pattern than was evident in the PLS regression analysis.
Comparison of hierarchical linear mixed-effects models showed that, for Pre-Meno females, adding Age to a model with Task Difficulty did not provide a significantly better fitting or a more parsimonious model (χ2 = 2.8, p = 0.24, ΔAIC = −1). For Post-Meno females, adding Age to a model with Task Difficulty significantly improved model fit and parsimony (χ2 = 10.6, p = 0.005, ΔAIC = −7). These results suggest that adding chronological age as a predictor provides a better explanatory model of the variability in the dependent variables for Post-Meno females, but not Pre-Meno females.
Post hoc linear regressions to explore education effects
Given that there was a trend in group differences in education, with Pre-Meno females trending toward being more educated than Post-Meno females (p = 0.055), we conducted four linear regressions models to explore whether Menopause Group (−1 Pre-Meno, +1 Post-Meno), standardized Education (zEDU) Group, and the Menopause Group * zEDU interaction predicted CS Easy, CS Hard, Recog Easy, and Recog Hard. There was no significant Menopause Group * zEDU interaction observed in any of the models (p-value range, 0.30–0.99). There was a significant effect of zEDU for both Menopause Groups for CS Hard accuracy (standardized β = 0.246, t = 2.17, p = 0.033).
Task fMRI results
The B-PLS analysis was conducted to determine how menopause status affected correlations between memory performance and brain activity at encoding and retrieval; and how this varied with chronological age within group. Performance was assessed using both spatial context and recognition accuracy to explore whether groups engaged similar or different brain regions to support these two different response types. The B-PLS fMRI analysis identified two significant LVs. The post hoc between-menopause group, repeated-measures MANOVA to assess whether there was an effect of a female's history of antidepressant use, HBC use, and diagnosis of PCOS on PLS brain scores identified no significant main effects or interactions for the two significant LVs [p > 0.05, first LV (LV1) and LV2].
LV1: correlations between retrieval activity and memory performance in Pre-Meno females
LV1 accounted for 15.55% of cross-block covariance and was significant at p = 0.001 based on permutation testing. Figure 3A presents the brain–behavior correlation profile for LVI, with 95% CIs for the correlation effects based on the bootstrapping, and summarizes how chronological age, spatial context retrieval accuracy, and recognition accuracy correlated with brain activity in regions identified in the singular image (Fig. 3C) during EncEasy, EncHard, RetEasy, and RetHard conditions. If the CI crosses 0, the effect is not significant for that group and condition. Figure 3B shows the scatterplots of brain scores by age and performance for Pre-Meno and Post-Meno females. Table 2 presents the local maxima for LV1. This LV primarily identified positive salience brain regions, which included lateral PFC, angular gyrus/inferior parietal cortex, and midline cortical regions i.e., medial PFC (mPFC), anterior cingulate, posterior cingulate, and precuneus.
Local maxima of negative and positive saliences identified in LV 1
Brain–behavior PLS results—LV1. A, The brain–behavior correlation profiles for LV1 depicting how chronological age, spatial context retrieval accuracy, and item retrieval accuracy correlated with activity in regions identified in the singular image. The bars represent the correlation values among age, spatial context retrieval accuracy, recognition accuracy and the brain activity pattern identified by the symmetrically paired singular image. The whiskers on the bars reflect the 95% CIs based on the bootstrap results. Bars in which the whiskers cross 0 indicate that correlation effect did not significantly contribute to the results. B, Scatterplots that depict the correlations between subjects' brain scores and age, spatial context retrieval accuracy, and recognition accuracy for Pre-Meno females (in yellow) and Post-Meno females (in blue). The slopes of these correlations, with 95% CIs, represent the bar values and whiskers in A, respectively. C, The singular image for LV1, threshold bootstrap ratio of ±3.28 (p < 0.001). The figure highlights regions identified between TR/lags 3–5 based on the temporal brain scores. The scale color codes are affected by the strength of BSR. Interpretation of this LV is presented in the Results section. The singular image is presented on template images of the lateral and medial surfaces of the left and right hemispheres of the brain using Caret software version 5.65 (https://neuro.debian.net/pkgs/caret.html).
Overall, Pre-Meno females accounted for LV1 effects. In Pre-Meno females, increased activity in positive salience regions during EncEasy, EncHard, RetEasy, and RetHard conditions was correlated with better spatial context memory. Importantly, retrieval effects were only observed in Pre-Meno females. Pre-Meno females who exhibited more retrieval activity in positive salience regions also had higher spatial context memory accuracy and lower recognition accuracy across Easy and Hard conditions. In addition, in Pre-Meno females age was correlated with less retrieval activity in these brain regions during the EncHard and RetHard conditions.
Although LV1 primarily reflected correlations between retrieval activity and memory performance in Pre-Meno females, Figure 3, A and B, indicates that Post-Meno females who exhibited more activity in positive salience regions during encoding had lower subsequent spatial context memory accuracy and higher subsequent recognition accuracy. However, this effect was only observed in the EncHard conditions. Also, only in Post-Meno females was this effect correlated with age. Older Post-Meno females exhibited more activity in these regions during EncHard, and this was linked to lower subsequent spatial context accuracy and higher subsequent recognition for the Hard condition.
LV2: menopause group differences in age-related and performance-related patterns of brain activity
LV2 (p = 0.006) accounted for 10.59% of cross-block covariance (Fig. 4, Table 3). This LV primarily identified negative salience brain regions that included: ventral occipitotemporal cortex and, bilateral parahippocampal gyrus. The correlation profile and brain–behavior scatterplots (Fig. 4A,B) indicate that LV2 identified group differences in age-related and performance-related brain activity. Post-Meno females contributed more strongly to this LV. Specifically, Post-Meno females who exhibited more activity in negative salience brain regions during encoding and retrieval also had higher spatial context retrieval accuracy and lower recognition accuracy. However, with advanced age within Post-Meno females, there was less activity in negative salience regions, and thus lower spatial context accuracy and higher recognition accuracy.
Local maxima of negative and positive saliences identified in LV 2
Brain–behavior PLS results—LV2. A, The brain–behavior correlation profiles for LV2 depicting how chronological age, spatial context retrieval accuracy, and item retrieval accuracy correlated with activity in regions identified in the singular image. B, Scatterplots that depict the correlations between subjects' brain scores and age, spatial context retrieval accuracy, and recognition accuracy for Pre-Meno females (in yellow) and Post-Meno females (in blue). The slopes of these correlations, with 95% CIs, represent the bar values and whiskers in A, respectively. C, The singular image for LV2, threshold bootstrap ratio of ±3.28 (p < 0.001). The figure highlights regions identified between TR/lags 3–5 based on the temporal brain scores. The scale color codes are affected by the strength of BSR. The singular image is presented on template images of the lateral and medial surfaces of the left and right hemispheres of the brain using Caret software version 5.65 (https://neuro.debian.net/pkgs/caret.html). Interpretation of this LV is presented in the Results section.
The effects in Pre-Meno females were in the opposite direction to those observed in Post-Meno females and were more complex. Pre-Meno females who exhibited less activity in these regions during EncEasy had lower spatial context retrieval accuracy and higher recognition accuracy. Advanced age in Pre-Meno females was correlated with more activity in negative salience brain regions during RetEasy conditions only, but was not correlated with performance. Thus, LV2 identified Menopause Group differences in age-related and performance-related patterns of brain activity.
Discussion
In the current study, we used multivariate behavioral and functional neuroimaging analysis methods to examine how the menopause status affected episodic memory for face–location association (spatial context memory) and its functional neural correlates, and how this correlated with age and memory performance within group. The behavioral PLS regression analysis indicated that both older age and Post-Meno, compared with Pre-Meno, status was related to lower spatial context retrieval accuracy and higher recognition accuracy. This result is consistent with past work that has indicated that Post-Meno females perform poorer on associative memory tasks compared with age-matched Pre-Meno females (Rentz et al., 2017) and that context memory accuracy is lower at midlife compared with young adulthood (Kwon et al., 2016). When probed within groups, however, we found that advanced age was associated with poorer spatial context retrieval accuracy in Post-Meno, but not Pre-Meno, females. This suggests that the detrimental role of age on spatial context memory emerges in late midlife, at a time when females with ovaries have already transitioned through menopause. Overall, our findings show that older Post-Meno females experience deficits in spatial contextual recollection, in the presence of intact familiarity-based item retrieval and/or item-specific recollection (Yonelinas et al., 1996).
The multivariate task fMRI analysis identified two significant LVs. LV1 identified differences in the correlations among age, episodic retrieval accuracy, and brain activity during encoding and retrieval. LV2 also identified group differences in the correlations between brain activity, age, and memory performance, which corroborates our hypothesis that at a neural level there are age-related differences in brain activity at post-menopause status, that are not apparent at pre-menopause. Below we discuss our task fMRI findings in greater detail.
Group similarities in subsequent memory effects
LV1 identified significant correlations between encoding activity in lateral PFC, angular gyrus/inferior parietal cortex, and midline cortical regions, and subsequent memory performance in both Pre-Meno and Post-Meno females. Specifically, Pre-Meno females who exhibited more activity in these regions at encoding also exhibited better spatial context accuracy (and lower recognition accuracy). The midline cortical regions identified in this LV (i.e., medial PFC, cingulate, and precuneus regions) are part of the default mode network (Biswal et al., 1997; Greicius et al., 2003; Power et al., 2010). Prior episodic memory studies have found that greater activity in these regions is associated with successful episodic encoding and contextual recollection (Kim, 2010; Rugg and Vilberg, 2013; de Chastelaine et al., 2016). Similarly, increased activity in lateral PFC activity has been consistently observed during spatial context memory tasks at encoding and retrieval (Rajah et al., 2008, 2010; Mitchell and Johnson, 2009), and increased left angular gyrus activity has been observed during associative object–context encoding and recollection (Maillet and Rajah, 2014a; Branzi et al., 2021; Bellana et al., 2023). In addition, Jacobs et al. (2016) reported no significant group differences in lateral PFC activity during a verbal episodic encoding tasks between Pre-Meno and Post-Meno females, consistent with our current findings.
Interestingly, in the current study Post-Meno females exhibited subsequent memory effects in lateral PFC, angular gyrus and midline cortical regions during the EncHard condition. Moreover, in Post-Meno females, the subsequent memory effects during EncHard also correlated with chronological age. Older Post-Meno females exhibited less activity in the mentioned brain regions, and this was correlated with poorer subsequent spatial context memory. Given that the difficulty manipulation in the current study was one of encoding load, our results show that older Post-Meno females displayed a pattern of brain activity consistent with the prediction of the Compensation Related Utilization of Neural Circuits Hypothesis (CRUNCH): older Post-Meno females activated similar regions as younger Pre-Meno females, but at higher level of task demand, which may reflect compensation for neural inefficiencies (Cappell et al., 2010).
Group differences in age-related and performance-related brain activity
LV1 and LV2 identified group differences in how chronological age affected memory-related brain activity and corroborates our hypothesis that at a neural level there are menopause-related differences in the effect of age on memory-related brain activity. For example, in LV1, group differences were observed at retrieval. Pre-Meno females who exhibited more activity in lateral PFC, angular gyrus, and midline cortical activity during Easy and Hard retrieval conditions also had higher spatial context memory accuracy and lower recognition accuracy. Therefore, only in Pre-Meno females was reinstatement of encoding-related activity in lateral PFC, angular gyrus, and midline cortical regions at retrieval correlated with better spatial context memory. Moreover, age-related declines in this reinstatement during the more challenging RetHard conditions was detrimental to spatial context memory in only Pre-Meno females.
Past work has shown how reinstatement of encoding-related activity at retrieval is important for successfully remembering episodic memories (Gordon et al., 2014; Leiker and Johnson, 2015; Thakral et al., 2015; Hill et al., 2021). Different patterns of cortical reinstatement may support successful recollection and context retrieval, compared with successful recognition (Johnson et al., 2009). In a previous fMRI study, Thakral et al. (2015) reported that in young adults, cortical reinstatement of the same regions identified in LV1 (lateral PFC, angular gyrus, midline cortical regions) supported successful retrieval of context information. Johnson et al. (2009) reported that reinstatement of posterior midline cortical activity was particularly relevant for successful recollection in a young adult sample. Therefore, our results indicate that middle-aged Pre-Meno females engage brain regions similar to young adults to support successful spatial context memory.
In contrast, middle-aged Post-Meno females did not show cortical reinstatement of encoding-related activity in lateral PFC, angular gyrus, and midline cortical at retrieval. Importantly, there was also no significant correlation between chronological age and LV1 retrieval activity in Post-Meno females. However, LV2 results show that Post-Meno females can engage in cortical reinstatement; but the regions they reactivated at retrieval were distinct from those engaged by Pre-Meno females to support spatial context memory. Specifically, LV2 shows that Post-Meno females who exhibited greater activity in bilateral occipitotemporal and parahippocampal cortices at encoding and retrieval had higher spatial context memory accuracy and lower recognition memory accuracy. Interestingly, Johnson et al. (2009) found that cortical reinstatement effects in occipitotemporal regions supported both detailed recollection and more general recognition memory. However, increased activity in parahippocampal and inferior parietal cortices has explicitly been observed during spatial context memory tasks (Hayes et al., 2007; Diana et al., 2013; Ankudowich et al., 2016; Snytte et al., 2022). LV2 PLS results also indicated that older Post-Meno females exhibited lower levels of bilateral occipitotemporal and parahippocampal activity at encoding and retrieval, which was correlated with lower spatial context memory and higher recognition memory. Together, LV1 and LV2 results suggest that spatial context memory reductions in Post-Meno females compared with Pre-Meno females may in part be because of menopause/endocrine aging affecting the ability to reactivate lateral PFC, midline cortical, and angular gyrus at retrieval (LV1; Morcom, 2014; Vilberg and Rugg, 2014; Wing et al., 2015; Bellana et al., 2023); and, chronological age-related reductions in occipitotemporal, parahippocampal, and inferior parietal activity at encoding and retrieval. These results highlight the combined influences of menopause and chronological aging on episodic memory and related brain function in middle-aged Post-Meno females.
LV2 also showed that middle-aged Pre-Meno females exhibited correlations among age, performance, and brain activity in occipitotemporal, parahippocampal, and inferior parietal cortices at encoding and retrieval. These effects were not as strong as those observed in Post-Meno females, and they were in the opposite direction to those associations seen in Post-Meno females. Together with the retrieval effects observed in LV1, our results indicate that better spatial context memory performance at the Pre-Meno stage was observed in females who exhibited increased lateral PFC, angular gyrus, and midline cortical activity at encoding and retrieval, and decreased parahippocampal, dorsal occipitotemporal, and inferior parietal cortex activity.
In conclusion, we found that both menopause status and chronological age affect spatial context memory behaviorally, though the effect of chronological age is most evident at postmenopause. Menopause status directly affected the direction of age-related and performance-related correlations with brain activity in parahippocampal, and occipitotemporal cortices across encoding and retrieval. Moreover, we found that only Pre-Meno females exhibited cortical reinstatement of encoding-related activity in midline cortical, prefrontal, and angular gyrus/inferior parietal cortex, at retrieval. This suggests that spatial context memory abilities may rely on distinct brain systems at premenopause compared with postmenopause. Future studies should focus on exploring potential mechanisms by which menopause affects age-related differences in memory and brain function. For example, menopause is characterized by a marked decline in centrally circulating E2 (Foster, 2012; Harlow et al., 2012; He et al., 2021). Reduced central E2 in ovariectomized rats has been associated with deficits on spatial memory tasks, synaptic loss in the hippocampus and increased neuroinflammation (Au et al., 2016). Thus, future work exploring the links between blood-based measures of peripheral inflammation, hippocampal volumes, and memory-related brain function may advance our understanding of the underlying mechanisms contributing to menopause and age effects on memory-related brain function.
Footnotes
This work was supported by Canada Research Chairs Program Grant CRC-2022-00240, Canadian Institute of Health Research Sex and Gender Science Chair Grant GS9-171369, and Natural Sciences and Engineering Research Council of Canada Discovery Grant RGPIN-2018-05761, awarded to M.N.R. We thank the research participants of the Brain Health at Midlife and Menopause (BHAMM) Study for their time and contribution to science. We thank part-time research assistants (H. Azizi, R. Young, A. Condescu, L. Khyatt) and trainees (S. Subramaniapillai, G. Velez Largo, J. Kearley, A. Duval, J. Snytte, S. Loparco) who assisted in participant recruitment or testing or MRI quality control for support. In addition, we thank Teams 9 and 10, Women, Sex, Gender, and Dementia Theme of the Canadian Consortium on Neurodegeneration in Aging, the MRI Staff at the Cerebral Imaging Center at the Douglas Mental Health University Institute, and Dr. D. Cohen for help with recruitment.
The authors declare no competing financial interests.
- Correspondence should be addressed to M. Natasha Rajah at mnrajah{at}gmail.com.
