Sex Differences in the Neurovascular Health of Aging Adults

Abstract

Background:

Poor cerebrovascular reactivity is associated with higher risk of cerebrovascular disease. The most common method to study cerebrovascular reactivity in aging adults, transcranial Doppler (TCD) ultrasound, yields measurements in large intracranial arteries, but not in regional brain parenchyma that may be more impaired in some disease processes. Measurements derived with TCD ultrasound suggest that there are sex differences in cerebrovascular reactivity for aging adults. We investigated the association between age and sex on cerebrovascular reactivity using blood oxygen-level dependent (BOLD) MRI in a representative group of aging adults.

Methods:

This cross-sectional study investigated BOLD cerebrovascular reactivity to CO₂ in a representative group of aging adults 51–83 years old. We manipulated ETCO₂ with breathing exercises and evaluated changes in 6 brain regions: whole brain, white matter, cortical gray matter, subcortical gray matter, left hippocampus, and right hippocampus. We used one linear regression per region to investigate the effects of age, sex, and their interaction on BOLD cerebrovascular reactivity.

Results:

We report an age-by-sex interaction for all brain regions (p≤0.050), except cortical grey matter (p=0.062). For white matter and subcortical gray matter, female participants trended toward an age-related BOLD cerebrovascular reactivity increase (p≤0.058), while male participants did not change with age (p>0.580). In the whole brain and bilateral hippocampi, the age trends for each sex were in opposite directions but not significant (p>0.211). We report a main effect of sex (female greater than male participants) for subcortical gray matter and the right hippocampus (p≤0.048) and no main effect of age in any model.

Conclusions:

We present the first report of age-related BOLD cerebrovascular reactivity increases in older female participants and higher BOLD cerebrovascular reactivity in older female compared to male participants. Sex and age-by-sex-based differences appear to be driven by changes in white matter, subcortical gray matter, and bilateral hippocampi.

Graphical abstract

Background

Cerebrovascular disease is a leading cause of long-term disability, high healthcare costs, and mortality.^1,2 Cerebrovascular disease is also an early predictor of age-related cognitive decline,³ Parkinson’s disease,⁴ multiple sclerosis,⁵ and Alzheimer’s disease.⁶ An effective method to assess cerebrovascular health is cerebrovascular reactivity (CVR) to CO₂, which measures the brain’s ability to regulate blood flow in response to changes in the arterial partial pressure of CO₂ (P_aCO₂) and, subsequently, blood pH. Different CVR to CO₂ methodologies use alternative approaches to measure blood flow (blood oxygen level dependent - BOLD, transcranial Doppler ultrasound – TCD, extracranial artery ultrasound, arterial spin labeling – ASL, phase contrast, positron emission tomography) or alter P_aCO₂ (gas inhalation, breath-hold, rebreathing, acetazolamide).

Most studies report that CVR to CO₂ decreases with age.⁷ Age-related BOLD-CVR decreases are reported in the frontal white matter watershed areas⁸ and the temporal lobes.^9,10 Lower ASL-CVR in the hippocampi and cortical tissue of aging adults is associated with higher cardiovascular risk as measured by the Framingham cardiovascular risk profile, and ASL-CVR decreases are greater in those with mild cognitive impairment compared to healthy adults.¹¹ These declines in CVR to CO₂ are noteworthy because of associations between cognitive performance and BOLD-CVR in the whole brain (language),¹² temporal lobe (memory and attention), and hippocampus (memory).⁹ Therefore, BOLD-CVR changes in these areas may be an early biomarker for cerebrovascular disease and cognitive decline.

Some studies report higher BOLD-CVR in male versus female participants,^13,14 whereas other studies report no sex differences at all.^8,15,16 However, most studies that investigate the effects of sex and age on vascular health use TCD-CVR. Lower CVR values in female than male participants are also reported for investigations of aging with TCD-CVR to CO₂.^17,18 Early studies suggest that TCD-CVR to CO₂ of the middle cerebral artery decreases in after menopause compared to premenopause,^17,18 even when compared with female participants of the same age,¹⁸ TCD-CVR to CO₂ is also decreased in older, postmenopausal female compared to younger and male participants of the same age.¹⁸ Lower CVR post menopause is hypothesized to be associated with decreases in circulating estrogen.^17,18

Despite correlations between MRI- and TCD-based CVR values,¹⁹ there are some limitations to TCD-CVR to CO₂ assessments. TCD-CVR is sensitive to miss-measurement of blood vessel diameters while the reliability of BOLD-CVR is not dependent on measuring vessel diameters. TCD-CVR measures blood flow velocity changes in large intracranial arteries, which is not a direct measurement of venule and capillary response while BOLD-CVR is directly sensitive to blood flow changes in the parenchymal vasculature. Finally, TCD-CVR assessment is limited to 1 or 2 major intracranial vessels simultaneously while BOLD-CVR can measure from the entire brain at once. Thus, the effects of age, sex, and their interaction on BOLD-CVR in the current study may not reflect findings seen in previous TCD-CVR studies.²⁰ This study uses BOLD-CVR to CO₂ to investigate age- and sex-based differences in a representative ageing population. We hypothesize that BOLD-CVR will decrease overall with age, particularly in the hippocampus as it is associated with cognitive decline. We also hypothesize that BOLD-CVR decreases will be more profound in older female participants, because they will primarily be postmenopausal, which is associated with lower TCD-CVR to CO₂.

Methods

Participants

We recruited participants from the Healthy Aging Brain Study: Health Disparities (HABS-HD) (IRB: 2016–128) into this study (Mechanisms of Aortic arch Stiffness and Brain Insult (MASBI) (IRB: 2021–040). Briefly, eligibility criteria for the HABS-HD study includes: adults age 50 or older who identify as Mexican American, African American, or non-Hispanic White, willing to give blood samples, able to undergo neuroimaging, and fluent in English or Spanish. Exclusion criteria include: type 1 diabetes, active infection, current or recent cancer (other than skin), current or recent traumatic brain injury with loss of consciousness, current severe mental illness (except depression), current or recent alcohol or substance abuse, active severe medical condition that can impact cognition, or current diagnosis of dementia other than Alzheimer’s disease.²¹ MASBI is an observational cross-sectional study that recruited participants from the HABS-HD project from March to August 2022 based on their prior permission to contact them for future studies. This study is a primary analysis for the MASBI project. Both studies were approved by the North Texas Regional Institutional Review Board, and were conducted at the University of North Texas Health Science Center (UNTHSC). All study procedures and risks were described to the participants before they provided written informed consent for each study. We used STROBE reporting guidelines to report methods.²²

Data Availability

Data from the HABS-HD Study are available to qualified scientists through the HABS-HD Study Data Access and Publications Committee. The BOLD and CO₂ data for the MASBI Study are available pursuant to institutional policies upon reasonable request.

MRI Procedures

Image Acquisition

All participants underwent multiple imaging sessions at the Institute for Translational Sciences Imaging Center at UNTHSC. We used Siemens 3T MAGNETOM Skyra and Vida magnetic resonance imaging (MRI) scanners (Siemens, Erlangen, Germany) and a 20-channel head/neck coil. The HABS-HD imaging sessions yielded high-resolution T1-weighted images and the MASBI imaging session yielded BOLD images (150 volumes; 8.5 minutes). Scan parameters are in listed in Supplemental Methods. The HABS-HD study collects T1-weighted images at multiple study sessions, thus we selected the T1-weighted image closest in time to the MASBI study imaging session.

Breathing Tasks to Create a Vasoactive Stimulus

We used breathing tasks to modify the P_aCO₂ as a vasoactive stimulus. During the BOLD scan, we played the breathing instructions to the participant through the main bore speaker of the MRI scanner. These instructions cued the participant to perform 8 tasks with 1 minute between the start of each task: 2 breath-holds (5-second inspiration, up-to 20 seconds of breath hold, and expiration) and then 6 sighs (a 3-second inspiration and a 3-second expiration). The breathing instructions were available in either English or Spanish, depending on participant preference. Participants completed a coaching session for the breathing tasks before entering the MRI scanner.

End Tidal Carbon Dioxide Acquisition and Processing

We recorded participants’ exhaled CO₂ with a capnograph (Respsense II, Nonin Medical Inc, Plymouth, Minnesota, USA). We placed an oral/nasal cannula on the participant, and attached a 10-meter sampling line that was connected to the CO₂ monitor outside the MRI room. Raw CO₂ data were sampled at 4 Hz, saved, and exported to files for processing. We visually inspected the raw exhaled CO₂ data, evaluated the measurement quality, and selected the appropriate segment of data for processing. From the selected data, we used a custom algorithm to identify CO₂ peaks that define end tidal CO₂ (ETCO₂). We then fitted the peaks with a spline interpolation and resampled the ETCO₂ interpolation to match the sampling frequency of the BOLD time points. We performed all CO₂ signal processing analyses in MATLAB R2023a (Mathworks, Natick MA).

We exported the ETCO₂ timeseries’ to R Statistics to calculate the root mean square of ETCO₂ for each participant for the full BOLD-CVR scan. We also exported ETCO₂ timeseries’ of a baseline period and post-baseline period to compare the median ETCO₂ and ETCO₂ variance for each period by sex or age. Details of how the baseline and post baseline periods were determined are in Supplemental Methods. We compared the ETCO₂ root mean square of the full BOLD-CVR scan by age and sex to determine if absolute ETCO₂ changes differed.

T1-Weighted Image Processing

We skullstripped T1 images with the Computational Anatomy Toolbox (CAT12) segmentation module²⁴ in Statistical Parametric Mapping (SPM12 )²⁵ and registered them to the BOLD images using the FMRIB software library image registration tool (FSL FLIRT)^26,27 (see Supplemental Methods for details).

BOLD Image Processing

Our BOLD pipeline is written in MATLAB R2023a (Mathworks, Natick, MA) and uses the FSL toolbox. (1) The BOLD data were converted from dicom to nifti (dcm2niix)²⁸ and imported into MATLAB. (2) We deleted the first 3 BOLD volumes to account for T1 saturation effects. (3) We motion corrected the BOLD volumes with a two-step process. First, we used a rigid motion correction (FSL McFLIRT) to register all BOLD volumes to the raw temporal mean.²⁷ Second, we used a nonrigid motion correction (MATLAB) to register all BOLD data to the revised temporal mean. (4) We identified BOLD volumes with unexpectedly high global intensity values (values greater than the 75^th percentile plus 1.5 times interquartile range for the participant) using derivative of root mean square variance over voxels (DVARS) for each volume. (5) We used standard FSL FEAT processing for slice-time correction, smoothing (5 mm gaussian kernel), grand mean scaling, and high-pass filtering (200 seconds).^29,30 (6) We used FSL to regress the 6 rigid motion parameters and censor (remove influence of) any BOLD volumes with a DVARS value higher than the participant’s threshold. We excluded participants for excessive BOLD motion if more than 10% of their data were censored due to high DVARS values, more than 10% of their data had >0.8 mm relative displacement between consecutive volumes, or more than 5 volumes had >3 mm absolute displacement from the starting volume.

We coregistered the BOLD images to the participants’ T1-scan with FSL FLIRT using boundary-based registration with the skull-stripped brain from CAT12. We normalized the participants’ to Montreal Neurological Institute (MNI) 152 space (2 mm³ isotropic voxels) using a non-linear registration with a 10 mm warp resolution.³¹ We visually inspected each coregistration for accuracy. We used the generated warps to transform the participants’ T1 brain masks into subject BOLD space.

CVR to CO₂ Calculation

We calculated whole-brain CVR to CO₂ in the frequency-domain using our coherence-weighted general linear model algorithm.³² Calculation in the frequency domain eliminates the need to synchronize the end-tidal CO₂ and BOLD data; coherence-weighting favors frequencies with high signal to noise ratios that are present in both end-tidal CO₂ and BOLD data. We generated the CVR to CO₂ maps in participant BOLD-space and masked them with their transformed T1 brain mask.

Anatomical Coregisration

We coregistered the CVR maps to MNI152 space for group comparisons. We used the Harvard-Oxford Subcortical and Cortical Atlas in FSL and extracted the median CVR for each participant from 6 segments: whole brain (excludes cerebellum and ventricles), cortical gray matter, subcortical gray matter (excluding cerebellum), left hippocampus, right hippocampus, and white matter. The hippocampi were chosen due to association with dementias and cognitive decline.^9,11 We used median values and nonparametric methods to limit the influence of outlier values in each region of interest.

Demographic, Clinical, Medication, Bloodwork, Questionnaire, and Cognitive Covariates

The HABS-HD study collected all covariate values at the same timepoint as the T1 imaging.²¹ We included the following covariate types: demographic (e.g. age, income, race, ethnicity, education), clinical (e.g. disease status), medications, bloodwork, questionnaires, and cognitive tasks. We calculated the Charlson Comorbidity Index retrospectively, as described in Supplemental Methods. Medication identification is described in Supplemental Methods.

The MASBI study collected additional covariate values at the same timepoint as the BOLD-CVR imaging. We included the following covariate types: respiration rate (from ETCO₂ monitoring during the BOLD-CVR imaging), blood pressure and heart rate (from a tonometry evaluation before or after the MRI), as described in Supplemental Methods.

Statistical Analysis

We used R Statistics version 4.3.2 to conduct analyses. We summarized demographic and clinical characteristics with medians and inter-quartile ranges (IQR) for continuous variables and counts and percentages for categorical variables. We tested sex differences with Kruskal-Wallis tests for continuous variables, and either Chi-Square or Fisher Exact tests for categorical variables. If a participant was missing a demographic or clinical datapoint, then we excluded them from comparisons using that datapoint but we retained them for other analyses. All participants had age and sex data. To investigate the main effects of age and sex as well as their interaction on CVR, we performed 1 linear regression for each of the 6 atlas segments with the following equation:

$$Y={\beta }_0+{\beta }_{1}Age+{\beta }_{2}Se{x}_{Female}+{\beta }_3\left.\left(Age\times Se{x}_{Female}\right.\right)$$ #(1)

To address potential sources of bias, demographic and clinical characteristics that differed by age or sex were considered for inclusion as confound variables in the linear regressions. When an interaction between age and sex was present, we investigated interaction effects using tests of marginal effects of trends with the emmeans package.³⁴ Exact P values are presented for all comparisons where p>0.001.

Results

Participants

From the HABS-HD study, 66 participants (age 51–83 years) were recruited into the MASBI study. We excluded N=13 participants from final analysis: N=1 missing data, N=5 excess motion in BOLD data, N=2 low quality CO₂ data, N=5 prior stroke or heart attack. We did not exclude other health conditions other than those excluded based on the eligibility criteria. Table 1 summarizes the characteristics of the remaining N=53 participants (30 female, 23 male). Table 2 summarizes physiological measures. Additional extensive participant description is in Tables S1–S9 and Figures S1–S9. Participant characteristics did not vary by sex (p>0.060) except for two factors: whether they were retired (p=0.005) and their Digit-Symbol Substitution Test z-score (p=0.007). Household income (p=0.078) and marital status (p=0.060) sex-differences were also close to significant. Participant characteristics did not vary by age (Table S1, Figs. S1–S6), except retirement status (p<0.001), where retired participants were older, as expected (Table S1). The Charlson Comorbidity Index was positively correlated with age due to age-related points, rather than illness-related points (Figure S2, p<0.001). We observed anticipated age- and sex-based differences, including age-based reductions in hemoglobin (Figure S4, p=0.024), diastolic blood pressure (Figure S5, p=0.012), and heart rate (Figure S5, p=0.036), and sex-based differences (Table 1) where male participants had higher levels of creatinine (p=0.003) and hemoglobin (p<0.001) and female participants had a higher respiration rate (Table 2, p=0.038).

Table 1.

Participant Characteristics by Sex

	Female (N=30)	Male (N=23)	Total (N=53)	P value
Age (years)				0.158‡
Median (Q1, Q3)	70 (64, 74)	65 (60, 72)	68 (62, 74)
Education (years)				0.750‡
Median (Q1, Q3)	14 (13, 16)	15 (12, 17)	14 (12, 16)
Household Income (United States Dollars)				0.078‡
Median (Q1, Q3)	55000 (15761, 75000)	73500 (59750, 112500)	65000 (24720, 80000)
Missing	4	1	5
Retired	22 (73.3%)	8 (34.8%)	30 (56.6%)	0.005§
Retirement Age (years)*				0.557‡
Median (Q1, Q3)	62 (55, 65)	60 (58, 66)	62 (56, 65)
Years Retired*				0.557‡
Median (Q1, Q3)	12 (9, 16)	10 (4, 16)	12 (7, 16)
Race/Ethnicity				0.339§
Black Non-Hispanic	13 (43.3%)	6 (26.1%)	19 (35.8%)
White Hispanic	10 (33.3%)	8 (34.8%)	18 (34.0%)
White Non-Hispanic	7 (23.3%)	9 (39.1%)	16 (30.2%)
Marital Status				0.060§
Married	12 (40.0%)	16 (69.6%)	28 (52.8%)
Divorced	9 (30.0%)	4 (17.4%)	13 (24.5%)
Separated	1 (3.3%)	1 (4.3%)	2 (3.8%)
Widowed	7 (23.3%)	0 (0.0%)	7 (13.2%)
Never Married	1 (3.3%)	2 (8.7%)	3 (5.7%)
High School Education				0.982§
Neither	4 (13.3%)	3 (13.0%)	7 (13.2%)
Diploma	25 (83.3%)	19 (82.6%)	44 (83.0%)
General Equivalency Diploma (GED)	1 (3.3%)	1 (4.3%)	2 (3.8%)
Type 2 Diabetes	6 (20.0%)	7 (30.4%)	13 (24.5%)	0.382§
Dyslipidemia	20 (66.7%)	18 (78.3%)	38 (71.7%)	0.353§
Hypertension	21 (70.0%)	16 (69.6%)	37 (69.8%)	0.973§
Hypothyroid	4 (13.3%)	4 (17.4%)	8 (15.1%)	0.715||
Anemia	8 (26.7%)	4 (17.4%)	12 (22.6%)	0.519||
Cardiovascular Disease	2 (6.7%)	3 (13.0%)	5 (9.4%)	0.642||
Depression	10 (33.3%)	3 (13.0%)	13 (24.5%)	0.115||
Mild Cognitive Impairment †	8 (26.7%)	9 (39.1%)	17 (32.1%)	0.335§
Smoker				0.115||
Never	21 (70.0%)	10 (43.5%)	31 (58.5%)
Past	8 (26.7%)	12 (52.2%)	20 (37.7%)
Currently	1 (3.3%)	1 (4.3%)	2 (3.8%)
Alcohol Use Disorders Identification Test (AUDIT)				0.811‡
Median (Q1, Q3)	1 (0, 3)	2 (0, 2)	2 (0, 2)
Subjective Memory Complaint Questionnaire (SMCQ)				0.125‡
Median (Q1, Q3)	4 (1, 7)	2 (0, 4)	3 (1, 6)
Missing	1	1	2
Rapid Assessment of Physical Activity (RAPA) Scale 1				0.446‡
Median (Q1, Q3)	5 (3, 6)	4 (3, 6)	4 (3, 6)
Rapid Assessment of Physical Activity (RAPA) Scale 2				0.648‡
None	12 (40.0%)	11 (47.8%)	23 (43.4%)	0.733||
Muscle Strength Activities	1 (3.3%)	1 (4.3%)	2 (3.8%)
Flexibility Activities	6 (20.0%)	6 (26.1%)	12 (22.6%)
Both	11 (36.7%)	5 (21.7%)	16 (30.2%)
Mini-Mental State Examination (MMSE)				0.341‡
Median (Q1, Q3)	28 (27, 30)	28 (26, 29)	28 (27, 29)
Charlson Comorbidity Index (CCI)				0.477‡
Median (Q1, Q3)	3.0 (2.0, 3.0)	2.0 (2.0, 3.0)	2.0 (2.0, 3.0)
Digit Span (z-score)				0.679‡
Median (Q1, Q3)	0 (−0, 1)	0 (−1, 1)	0 (−0, 1)
Digit-Symbol Substitution Test (z-score)				0.007‡
Median (Q1, Q3)	0.40 (−0.15, 0.70)	−0.40 (−0.95, 0.25)	0.10 (−0.50, 0.50)
Trail Making Test A (z-score)				0.928‡
Median (Q1, Q3)	0.25 (−0.55, 0.80)	0.30 (−0.40, 0.50)	0.30 (−0.40, 0.50)
Trail Making Test B (z-score)				0.879‡
Median (Q1, Q3)	0.05 (−1.05, 0.70)	0.10 (−0.85, 0.40)	0.10 (−1.10, 0.70)
Creatinine (mg/dL)				0.003‡
Median (Q1, Q3)	0.79 (0.69, 0.85)	1.04 (0.81, 1.19)	0.83 (0.72, 1.06)
Missing	1	0	1
Hemoglobin A1C % of Total Hemoglobin				0.161‡
Median (Q1, Q3)	5.6 (5.4, 6.0)	5.8 (5.6, 6.5)	5.7 (5.4, 6.2)
Missing	1	0	1
Hemoglobin (g/dL)				< 0.001‡
Median (Q1, Q3)	12.6 (11.6, 12.8)	14.3 (13.6, 15.0)	12.9 (12.4, 14.2)
Missing	1	0	1
Mean Cellular Hemoglobin Concentration (g/dL)				0.009‡
Median (Q1, Q3)	33.1 (32.4, 33.5)	33.8 (33.2, 34.3)	33.3 (32.7, 33.9)
Missing	1	0	1
Mean Cellular Volume (fL)				0.580‡
Median (Q1, Q3)	90.4 (84.4, 93.4)	87.8 (85.8, 92.1)	89.6 (85.2, 93.0)
Missing	1	0	1

Clinical and demographic information was collected at the same time as the T1 image, and age used for analyses was the participant age at the time of the blood oxygen level dependent (BOLD) scan. Disease states were determined by clinical consensus variables in the HABS-HD data (diabetes, dyslipidemia, hypertension, hypothyroid, anemia, cardiovascular disease, depression, mild cognitive impairment). Of the participants with missing household income, N=4 (3 female) were retired and did not list any other income, and N=1 (female) listed an occupation but not income information. The Charlson Comorbidity Index was retrospectively calculated based in interview medical history/questionnaires and reported medication. Sex differences in bloodwork were within the expected ranges for female and male participants and within normal ranges. Abbreviations: AUDIT – Alcohol Use Disorders Identification Test, CCI – Charlson Comorbidity Index, MMSE – Mini-Mental State Exam, RAPA – Rapid Assessment of Physical Activity (Scale 1 measures intensity levels and amount of physical activity, Scale 2 measures whether strength and flexibility activities are done weekly), SMCQ – Subjective Memory Complaint Questionnaire, mg/dL – milligrams per deciliter, g/dL – grams per deciliter, fL – femotoliters. P values are not adjusted for multiple comparisons. Legend:

^*Values are from participants that are retired only

^†no participants had dementia, as such, only mild cognitive impairment is included.

^‡Kruskal-Wallis rank sum test

^§Pearson’s Chi-squared test

^||Fisher’s Exact Test for Count Data.

Table 2.

Blood Pressure, Heart Rate, Respiration Rate, and ETCO₂ by Sex

	Female (N=30)	Male (N=23)	Total (N=53)	P value
	N, Median (Q1, Q3)	N, Median (Q1, Q3)	N, Median (Q1, Q3)
Systolic Blood Pressure, Wrist (mmHg)	N=28, 131 (120, 140)	N=23, 131 (114, 139)	N=51, 131 (116, 139)	0.443*
Systolic Blood Pressure, Carotid (mmHg)	N=28, 131 (123, 146)	N=23, 134 (114, 143)	N=51, 132 (119, 146)	0.272*
Diastolic Blood Pressure (mmHg)	N=28, 66 (60, 68)	N=23, 65 (58, 77)	N=51, 66 (59, 72)	0.820*
Diastolic Blood Pressure, Carotid (mmHg)	N=28, 66 (60, 68)	N=23, 65 (58, 77)	N=51, 66 (59, 72)	0.820*
Heart Rate (beats per minute)	N=28, 60 (55, 72)	N=23, 59 (56, 71)	N=51, 60 (55, 72)	0.970*
Respiration Rate (breaths per minute)	N=22, 16 (14, 19)	N=19, 14 (12, 16)	N=41, 15 (12, 18)	0.038*
Baseline Median ETCO2 (mmHg)	N=22, 34.11 (32.11, 35.68)	N=19, 33.95 (31.46, 37.5)	N=41, 34.06 (31.47, 36.06)	0.928†
Baseline ETCO2 Variance (mmHg)	N=22, 4.58 (1.33, 7.70)	N=19, 2.87 (0.77, 9.75)	N=41, 4.21 (1.20, 7.81)	0.651†
Post-Baseline Median ETCO2 (mmHg)	N=22, 35.73 (33.12, 37.72)	N=19, 36.52 (34.09, 41.14)	N=41, 36.10 (33.33, 38.45)	0.198†
Post-Baseline ETCO2 Variance (mmHg)	N=22, 9.41 (4.32, 25.27)	N=19, 7.11 (5.78, 12.14)	N=41, 7.60 (5.78, 19.79)	0.293†
Root Mean Square of ETCO2 (BOLD-CVR scan)	N=30, 35.84 (32.84, 37.52)	N=23, 34.95 (33.16, 39.45)	N=53, 35.69 (32.93, 38.11)	0.894†

N for each group is the N listed in table header unless otherwise noted. Blood pressure and heart rate are calculated from a tonometry session the same day as the BOLD-CVR scan. The Baseline and Post-baseline time periods are before the start of the ETCO₂ data used to calculate BOLD-CVR and during the period at the start of the BOLD-CVR exam. Respiration rate was calculated by counting the number of breaths during the 1-minute baseline period. See Supplemental Methods for details. Root Mean Square of ETCO₂ was calculated for the entire period of the BOLD-CVR scan (i.e. data used for BOLD-CVR calculation). Abbreviations: BOLD – blood oxygen level dependent, CVR – cerebrovascular reactivity. Legend:

^*Kruskal-Wallis rank sum test Wilcoxon

^†rank sum exact test. P values are not adjusted for multiple comparisons

Prescription medication use is reported by sex (Tables S2, S4, S5), age (Tables S3, S6), and disease state (Tables S7–S8). Number of total medications did not vary by age (Figure S6, p=0.471). Medication use was similar between female and male participants. Male participants were more likely to take gout medication. Medication use was largely independent of age, and in many cases few participants used a specific medication, thereby limiting analyses based on either age or sex. Only 5 female participants reported use of hormone replacement therapy, limiting analyses. Participants with hypertension, dyslipidemia, or type 2 diabetes reported more use of medications associated with their disease state (Tables S7–S8). Participants with type 2 diabetes reported more total medications than those without (Table S8).

Quality control metrics and in-scan behavior during the BOLD-CVR scan are reported in Table 2, Table S9 and Figures S7–S9. Head motion during the BOLD scans did not differ by sex (Table S9) or age (Figure S7) and was not associated with CVR values (Figure S9) (p>0.096). Baseline and post-baseline ETCO₂ did not differ by sex (Table 2) or age (Figure S8) (p>0.198). Mixed-effects models did not identify significant sex-by-timepoint or age-by-timepoint, nor a main effect of sex or age in their respective models (Supplemental Methods, p>0.930). Median ETCO₂ and ETCO₂ variance increased from baseline to post-baseline in the sex-by-timepoint models (Supplemental Methods, p<0.001). The root mean square of ETCO₂ during the BOLD-CVR scan did not vary by age (Figure S8) or sex (Table 2) (p>0.396).

BOLD-CVR Values

The grand median whole-brain BOLD-CVR map is shown in Figure 1. The results of the linear regression for each region are listed in Table 3. Figure 2 depicts median CVR versus age coded by sex. Overall, female participants trend toward increased CVR with age, while CVR declined in male participants, although the changes were only statistically significant for female participants and in white matter (p=0.046) and subcortical gray matter (p=0.049). However, if one introduces an age-by-sex interaction term into the model (Equation 1, Table 3), the age-by-sex relationship was significant in all brain regions except cortical gray matter (p=0.062), with female participants having higher CVR than male participants as they age. After controlling for this interaction, there was a residual sex difference (higher CVR in female participants) in subcortical gray matter and right hippocampus. The residual sex difference in subcortical gray matter and right hippocampus was of opposite direction to the age-by-sex relationship, blunting the overall sex difference. Cortical gray matter did not show any main effects, nor interactions.

Figure 1.

Grand Median of Blood Oxygen Level Dependent Cerebrovascular Reactivity to CO₂. This map (N=53) is presented in Montreal Neurological Institute (MNI) space, with z-axis slices at coordinates z = 61, 54, 44, and 40 (left to right). In these slices, we observe a clear demarcation between gray matter and white matter, evident in the more superior slices, z = 61 and 54. We also observe the lateral ventricles and deep gray structures in the more inferior slices, z = 44 and 40. The images in this figure were generated with MATLAB.

Table 3.

Linear Regression Model Values

	Estimate (Standard Error)				95% Confidence Interval				P value
Brain Region	Intercept	Sex	Age	Age × Sex	Intercept	Sex	Age	Age × Sex	Intercept	Sex	Age	Age × Sex
Whole Brain	0.0154 (0.0095)	−0.0249 (0.0134)	−0.0001 (0.0001)	0.0004 (0.0002)	(−3.75E-03, 3.45E-02)	(−5.18E-02, 1.94E-03)	(−4.34E-04, 1.43E-04)	(1.90E-07, 7.94E-04)	0.113	0.068	0.315	0.050*
White Matter	0.0091 (0.0069)	−0.0185 (0.0096)	−0.0001 (0.0001)	0.0003 (0.0001)	(−4.67E-03, 2.29E-02)	(−3.78E-02, 8.65E-04)	(−2.83E-04, 1.32E-04)	(8.68E-06, 5.80E-04)	0.190	0.061	0.468	0.044*
Cortical Gray Matter	0.0188 (0.0112)	−0.0279 (0.0157)	−0.0002 (0.0002)	0.0004 (0.0002)	(−3.72E-03, 4.14E-02)	(−5.96E-02, 3.69E-03)	(−5.21E-04, 1.57E-04)	(−2.25E-05, 9.12E-04)	0.100	0.082	0.286	0.062
Subcortical Gray Matter	0.0179 (0.0128)	−0.0364 (0.0179)	−0.0002 (0.0002)	0.0006 (0.0003)	(−7.76E-03, 4.36E-02)	(−7.24E-02, −3.08E-04)	(−5.68E-04, 2.06E-04)	(6.67E-05, 1.13E-03)	0.167	0.048*	0.352	0.028*
Left Hippocampus	0.0167 (0.0111)	−0.0291 (0.0155)	−0.0002 (0.0002)	0.0005 (0.0002)	(−5.56E-03, 3.90E-02)	(−6.04E-02, 2.12E-03)	(−5.13E-04, 1.58E-04)	(1.84E-05, 9.41E-04)	0.138	0.067	0.293	0.042*
Right Hippocampus	0.0238 (0.0119)	−0.0380 (0.0167)	−0.0003 (0.0002)	0.0006 (0.0002)	(−1.72E-04, 4.78E-02)	(−7.16E-02, −4.33E-03)	(−6.47E-04, 7.46E-05)	(1.19E-04, 1.11E-03)	0.052	0.028*	0.117	0.016*

Each model has N=53 participants (30 females, 23 males). P values within each linear regression are adjusted for the regression. P values are not adjusted across regressions.

Figure 2.

Scatterplots of age versus cerebrovascular reactivity by sex. Each panel depicts a region: (A) whole brain, (B) white matter, (C) cortical gray matter, (D) subcortical gray matter, (E) left hippocampus, and (F) right hippocampus. The y-axes are median cerebrovascular reactivity with units of the change in blood oxygen level dependent signal versus the change in end tidal CO₂ (ΔBOLD/ΔCO₂ mmHg), x-axes are age (years). Females (yellow, circle, N=30) and males (purple, triangle, N=23) differ in cerebrovascular reactivity with age. Females show increased CVR with age, particularly for white matter (B) and subcortical gray matter (D), and males show little change with age or a slight downward shift. Pearson R for female and male participants are listed for each brain region. Linear regression models (Table 3) indicate a sex-by-age interaction effect for all regions except cortical gray matter (C).

Tests of estimated marginal trends for the whole brain, left hippocampus, and right hippocampus did not show an age effect for either sex independent of the age-by-sex interaction. However, the age trends were in opposite directions for each region, females were positive (p>0.113) and males were negative (p>0.211). Tests comparing estimated marginal trends in white matter and subcortical gray matter revealed that females trended toward an increase with age (white matter: p=0.058, subcortical gray matter: p=0.051), but males did not change with age (white matter: p=0.717, subcortical gray matter: p=0.580).

Hemodynamic Response Function Alpha

The coherence-weighted general linear model we used to compute BOLD-CVR was validated in younger, healthier adults. The model includes the parameter, alpha, which describes the shape of the gamma function of the hemodynamic response function. Alpha was fixed at α=0.15 in our analyses, however, this value is potentially inappropriate for older adults or those with poor vascular health. To test if the age, sex, and age-by-sex effects seen were due to systematic differences in alpha, we used the basis-expansion method to estimate the temporal delay for BOLD to CO₂ changes for each participant’s whole brain segment. We performed a linear regression with the estimated alpha as the outcome variable and age, sex, and an age-by-sex interaction as the predictor variables. There were no systematic differences in estimated alpha value by age (b=−0.0009, SE=0.005, CI 95%: [−0.01, 0.009], p=0.857), sex (b=−0.43, SE=0.47, CI 95%: [1.38, 0.51], p=0.346), or age-by-sex (b=0.005, SE=0.007, CI 95%: [0.005, 0.007], p=0.452), thus, it is unlikely that the alpha value was a cause of the BOLD-CVR to CO₂ effects in this study.

Discussion

Our finding that BOLD-CVR in older female participants (51–83 years) increases with age is novel. Most BOLD-CVR studies report age-related decreases when comparing younger (e.g. ages 20s-40s) versus older (e.g. ages 50s-70s) adults,^7,9,15,19,35 or they report no age-related changes in grey matter.^7,16 However, one study also reported CVR decreases with age in cortical gray matter, cingulum, and the temporal lobe in adults 55–75 years old.⁹ Additionally, one longitudinal study reported that TCD-CVR decreased in older adults (70–80 years).³⁶ That decrease was correlated with subjective memory complaints, but not with sex. However, a pair of studies also report that TCD-CVR increased with age in response to a hypercapnic stimulus.^37,38 Those increases were not associated with either sex effects or with an age-by-sex interaction.³⁷

We report modest BOLD-CVR increases with age in female participants, and we report these increases in most brain subregions. TCD-CVR studies have associated menopause with greater decreases in CVR than those in males of corresponding ages.^17,18 Menopausal status was only available for a subset (N=14 of 30, 47%) of female participants, 93% (N=13 of 14) were postmenopausal. Thus, we assume female participants without a reported status are also predominantly postmenopausal. Prior studies report that higher estrogen concentrations may be a protective factor against age-related TCD-CVR reduction in females until menopause, when estrogen decreases.^17,18 They also report that hormone replacement therapy is associated with better CVR in the sixth decade of life and beyond.¹⁷ Furthermore, a later onset of menopause is associated with higher age-adjusted³⁹ and mean arterial pressure-adjusted⁴⁰ TCD-CVR. In our study, white matter BOLD-CVR increased in female but not in male participants with age. White matter CVR has been reported to be increased,⁴¹ unchanged,¹⁵ or decreased⁸ with age. These discrepant reports are unsurprising because BOLD-CVR signal is relatively low in white matter.

The hippocampi showed a bilateral age-by-sex effect, with a residual age effect in the right hippocampus only. Hippocampal BOLD-CVR is positively associated with cognitive ability in older (55–75 years) but not younger (21–44 years) adults.⁹ We report a residual independent sex effect in subcortical gray matter. Subcortical versus cortical gray matter are rarely investigated separately, but have been shown to have an attenuated ASL-CVR response to hypercapnic (but not hypocapnic) stimuli in older adults.⁴² Taken together, our findings indicate that, in aging adults, sex effects on BOLD-CVR occur primarily in subcortical, not cortical, gray matter. The whole brain effects are likely driven by white matter, bilateral hippocampi, and subcortical gray matter.

We did not find an age effect for any of the models. The lack of age effects is likely due to the age-by-sex interaction present in all models, except the cortical gray matter. Previous reports of the relationship between gray matter BOLD-CVR and age have been mixed, with some studies showing a decrease with age^9,10 and others showing no effect of age.^8,16,20 The lack of age effects in cortical gray matter in our study may be consistent with studies that do not report an effect in overall gray matter. The lack of age effect is possibly because we compared response within one older group (51–83 years), while many studies compare younger adults to older adults.

Because these data are cross-sectional, we cannot determine if increases in CVR to CO₂ are due to age, birth-year, or socio-cultural factors. A 3 year longitudinal study of adults (22–70 years) reported TCD-CVR values were stable and independent of age, sex, smoking, and follow-up interval.⁴³ However, a 4 year longitudinal BOLD-CVR study of adults (20–89 years) reported that BOLD-CVR decreased with time, with the most rapid decrease reported in 40–50 year-olds.¹⁰ Many studies investigating age and BOLD-CVR compare different age groups, rather than across an elderly age group. The age range in our study is 51–83 years. The greatest age-related BOLD-CVR decreases may have already occurred in the participants and increases associated with age in female participants are due to other factors. For example, female participants in our cohort might either have started with a higher BOLD-CVR earlier in life than male participants, but then had longitudinal decreases in BOLD-CVR, female participants may have experienced vascular protective factors that decreased in magnitude with birth year, or both. Female participants in our study were born between 1940 and 1970, a period with rapid generational changes in workplace demographic environment factors, such as less gender-based occupational segregation, increases of white women in workforces, and changes in job types (e.g. industrial versus office).⁴⁴ Workplace environment risk factors may have increased exposures and stressors that decreased CVR in female participants born later (i.e., the younger participants in our study), although this effect is speculative.

We explored other clinical, demographic, social determinants of health, and image quality characteristics that might explain our findings. Most of these factors did not differ by age or sex, or did not differ in a clinically meaningful or unexpected manner (e.g. values from blood work). The 2 sex-dependent differences in our participants were retirement status and Digit-Symbol Substitution Test score. The Digit-Symbol Substitution Test is sensitive, but non-specific, to deficits and changes in several domains including motor speed, associative learning, attention, and visuomotor skill.⁴⁵ However, including these 2 covariates in the models neither changed the effects, nor fit for any brain region.

Retirement status might have contributed to age-by-sex differences. If retirement is a proxy for socioeconomic status (e.g. lower lifetime of occupational risks or better healthcare access), then early retirement might improve BOLD-CVR and explain our age-by-sex differences. Both sexes had similar retirement ages, number of years retired, and household incomes, although many more female than male participants were retired, so these sex comparisons are limited.

Medications associated with age and sex differences or marginal trends had few participants with reported use, limiting the power of these results. However, corticosteroid (N = 7 participants, 6 female) and beta-blocker (N=12 participants, 5 female) use was associated with age (p=0.056 and 0.030, respectively). Exploratory inclusion of corticosteroids in the CVR models changed and reduced model effects, though the age-by-sex interaction remained (p<0.096) in all models. While corticosteroid use might contribute to the CVR variability, this should be interpreted with caution, since few participants report corticosteroid use, route of administration is not differentiated, and the corticosteroid models are underpowered. Exploratory models that included beta-blocker use did not change the age and sex effects.

Sex moderates the effect of aortic arterial stiffness on ASL-CVR in aging adults. Higher arterial stiffness (indexed by pulse wave velocity) is associated with higher ASL-CVR in aging males, but it is associated with lower ASL-CVR in aging (55–75 years) female participants.⁴⁶ This finding is interesting in the context of our results because it shows sex to have a differing relationship with another independent variable, and highlights the importance of not only accounting for sex in BOLD-CVR, but also investigating its potential interaction with other factors.

We did not use common clinical exclusion criteria (e.g. type 2 diabetes, cardiovascular disease, hypertension, neurological conditions), except to exclude individuals with a history of major vasculopathy (stroke or heart attack). In contrast, most studies investigating the effect of aging on CVR exclude individuals with clinical CVR risk factors (e.g. diabetes, cardiovascular disease). The heterogenous, yet representative, nature of our cohort possibly contributed to our findings, especially if risk factors associated with demographic and clinical characteristics change over time.

There were no sex differences for anemia and hemoglobin levels were within in normal ranges. Heavy menstrual bleeding is the most common cause of iron deficiency anemia in menstruating individuals.⁴⁷ After menopause, hemoglobin levels become on par with male individuals of the same age.⁴⁷ It is possible that the age-related increases in female participants occurred because of steady reversal of menstruation-related-anemia damage to blood vessels. Further research in a larger group can investigate individual differences in iron levels and anemia pattern with age and sex in relation to BOLD-CVR to CO₂.

Our BOLD-CVR maps are spatially consistent with maps reported by other studies, but the BOLD-CVR values we report are generally lower. Methodological differences in preprocessing and filtering the BOLD data might contribute to this difference. Motion is a major source of noise in BOLD data, and our processing pipeline includes 2 motion correction steps (rigid and non-rigid). In contrast, many BOLD pipelines use only 1 motion correction (rigid). We statistically regressed motion parameters in the BOLD data prior to calculating BOLD-CVR, which is not reported in all BOLD-CVR studies. Many studies do not quantify volunteer motion, nor report motion-based data rejection criteria. Furthermore, we also high-passed filtered our BOLD to correct for signal drift across the run, which is not performed in all BOLD-CVR studies. Each of these steps may account for lower CVR values, however each step is also a methodological improvement to BOLD-CVR preprocessing.

Limitations

This a cross-sectional study, not longitudinal, so it is not possible to determine if the results represent changes (or lack thereof) in any given individual. Our sample size for this study is larger than the group sizes for most BOLD-CVR studies. However, the study was not powered to investigate several individual differences directly in the regression models. While we performed secondary exploratory models containing variables of interest, future studies with larger sample sizes will benefit from including varied medications and disease statuses in omnibus models. Furthermore, demographic and clinical characteristics for the study sample were collected at the time of the T1 session, which was commonly a year before or after the BOLD-CVR session.

Most of our female participants are older and are likely postmenopausal, but we only have data on menstrual status for half of the participants, and do not have circulating hormone concentrations. Thus, we cannot ascertain whether our results in females are influenced by their hormones, menstruation, age, or birth year.

Medication use was self-reported and did not specify whether the medication was prescribed. We only included medications that are solely available by prescription in our analyses. As such, it is possible that we did not include some prescribed medications also available over-the-counter (e.g., analgesics and antihistamines). Blood pressure, heart rate, and respiration rate were measured for most participants on the same day as the BOLD-CVR scan. However, we do not have continuous measurements throughout the scan, and we do not have measures such as pulse oximetry, intracranial pressure, and cerebral profusion pressure, which also influence blood flow. Physiological state changes, which may impact CVR, may be induced by being in an MRI and time spent in the MRI. A follow-up study should evaluate these at the same time as the BOLD-CVR scan.

Our breathing tasks might induce transient hypercapnia during breath holds and hypocapnia during sighs.⁴⁹ Age-related ASL-CVR deficits might be greater during hypercapnia versus hypocapnia,⁴² thus we potentially reduce our signal due to our mixed CO₂ effects. ETCO₂ is an indirect measurement of arterial CO_2, and can only be measured during exhalation. As a result, rapid fluctuations in PaCO₂ occurring faster than the respiration rate will not be detected. Our breathing tasks may produce smaller end-tidal CO₂ changes versus gas-challenges. However, our root mean square ETCO₂ did not differ by age or sex. Additionally, others report that gas-challenge and breath-hold stimuli achieve similar results in CVR amplitude.⁵⁰ Breathing tasks might cause more head motion, but our intra-scan movement was not associated with sex, age, or BOLD-CVR for any region (Table S9, Figures S7, S9).

Conclusion

We are the first to report an age-related increase in BOLD-CVR to CO₂ in aging female participants, along with higher BOLD-CVR to CO₂ in aging female than male participants. The BOLD-CVR increases in our female participants were most prominent in subcortical gray matter, bilateral hippocampi, and white matter. Importantly, our participants represent a racially and ethnically diverse sample of aging adults. Although our findings are contrary to our hypotheses, these results highlight the importance of considering the interaction between sex and age when studying cerebrovascular health. Speculatively, it is possible that the increase in CVR to CO₂ in aging female participants occurred because menopause reduced cyclical anemia, therefor allowing some recovery of blood vessel reactivity.

Non-standard Abbreviations and Acronyms

ASL

arterial spin labeling

BOLD

blood oxygen level dependent

CAT12

computational anatomy toolbox

CVR

cerebrovascular reactivity

DVARS

derivative of root mean square variance over voxels

ETCO₂

end tidal carbon dioxide

FLIRT

FSL Image Registration Tool

FSL

FMRIB Software Library

GRAPPA

generalized autocalibrating partially parallel acquisitions

HABS-HD

Healthy Aging Brain Study: Health Disparities

IQR

inter-quartile range

MASBI

Mechanisms of Aortic arch Stiffness and Brain Insult

MNI

Montreal Neurological Institute

MRI

magnetic resonance imaging

MPRAGE

magnetization-prepared rapid acquisition gradient

McFLIRT

Motion Correction FMRIB Software Library Image Registration Tool

SPM

Statistical Parametric Mapping

TCD

transcranial doppler ultrasound

UNTHSC

University of North Texas Health Science Center